From 6f01b244a29e7b15b4a33168ceb4d603a6244c86 Mon Sep 17 00:00:00 2001 From: rattus <46076784+rattus128@users.noreply.github.com> Date: Wed, 10 Jun 2026 03:57:04 +1000 Subject: [PATCH 01/21] mm: dont reset cast buffers in cleanup_models_gc() (#14372) cleanup_models_gc can be called once per load_models_gpu via free_memory, which in turn can de-activate an active model via this reset_cast_buffers. cleanup_models_gc() could also come via obscure garbage collector paths so limit reset_cast_buffers to the post-node callsite instead. --- comfy/model_management.py | 2 -- 1 file changed, 2 deletions(-) diff --git a/comfy/model_management.py b/comfy/model_management.py index 8e786c0a5..9dc0a4e13 100644 --- a/comfy/model_management.py +++ b/comfy/model_management.py @@ -958,8 +958,6 @@ def loaded_models(only_currently_used=False): def cleanup_models_gc(): do_gc = False - reset_cast_buffers() - for i in range(len(current_loaded_models)): cur = current_loaded_models[i] if cur.is_dead(): From ad564899d37aec4cf151cf77b69bf1f4b6afe233 Mon Sep 17 00:00:00 2001 From: Kohaku-Blueleaf <59680068+KohakuBlueleaf@users.noreply.github.com> Date: Wed, 10 Jun 2026 03:55:29 +0800 Subject: [PATCH 02/21] Ensure conditions are not trainable to avoid bugs (#14368) --- comfy_extras/nodes_train.py | 6 ++++++ 1 file changed, 6 insertions(+) diff --git a/comfy_extras/nodes_train.py b/comfy_extras/nodes_train.py index 046eeaaf5..273f55e7c 100644 --- a/comfy_extras/nodes_train.py +++ b/comfy_extras/nodes_train.py @@ -15,6 +15,7 @@ import comfy.sampler_helpers import comfy.sd import comfy.utils import comfy.model_management +from comfy.conds import CONDRegular, CONDList from comfy.cli_args import args, PerformanceFeature import comfy_extras.nodes_custom_sampler import folder_paths @@ -120,6 +121,11 @@ def process_cond_list(d, prefix=""): process_cond_list(v, f"{prefix}.{k}") elif isinstance(v, torch.Tensor): d[k] = v.clone() + elif isinstance(v, CONDList): + v.cond = [t.detach() if isinstance(t, torch.Tensor) else t for t in v.cond] + elif isinstance(v, CONDRegular): + if isinstance(v.cond, torch.Tensor): + v.cond = v.cond.detach() elif isinstance(v, (list, tuple)): for index, item in enumerate(v): process_cond_list(item, f"{prefix}.{k}.{index}") From f8e51b674c75f41b3960d65ce83f77301ab297c9 Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?Jukka=20Sepp=C3=A4nen?= <40791699+kijai@users.noreply.github.com> Date: Wed, 10 Jun 2026 02:47:34 +0300 Subject: [PATCH 03/21] feat: Add Bernini-R model support (Wan video) (CORE-279) (#14216) --- comfy/ldm/wan/model.py | 31 ++++++++- comfy/model_base.py | 18 ++++++ comfy_extras/nodes_bernini.py | 115 ++++++++++++++++++++++++++++++++++ nodes.py | 1 + 4 files changed, 163 insertions(+), 2 deletions(-) create mode 100644 comfy_extras/nodes_bernini.py diff --git a/comfy/ldm/wan/model.py b/comfy/ldm/wan/model.py index 9178b3344..282408891 100644 --- a/comfy/ldm/wan/model.py +++ b/comfy/ldm/wan/model.py @@ -8,7 +8,7 @@ from einops import rearrange from comfy.ldm.modules.attention import optimized_attention from comfy.ldm.flux.layers import EmbedND -from comfy.ldm.flux.math import apply_rope1 +from comfy.ldm.flux.math import apply_rope1, rope import comfy.ldm.common_dit import comfy.model_management import comfy.patcher_extension @@ -570,6 +570,14 @@ class WanModel(torch.nn.Module): full_ref = self.ref_conv(full_ref).flatten(2).transpose(1, 2) x = torch.concat((full_ref, x), dim=1) + # In-context reference (Bernini) + context_latents = kwargs.get("context_latents", None) + main_len = x.shape[1] + if context_latents is not None: + for lat in context_latents: + cl = self.patch_embedding(lat.float().to(x.device)).to(x.dtype).flatten(2).transpose(1, 2) + x = torch.cat([x, cl], dim=1) + # context context = self.text_embedding(context) @@ -599,6 +607,9 @@ class WanModel(torch.nn.Module): # head x = self.head(x, e) + if context_latents is not None: + x = x[:, :main_len] + if full_ref is not None: x = x[:, full_ref.shape[1]:] @@ -606,7 +617,7 @@ class WanModel(torch.nn.Module): x = self.unpatchify(x, grid_sizes) return x - def rope_encode(self, t, h, w, t_start=0, steps_t=None, steps_h=None, steps_w=None, device=None, dtype=None, transformer_options={}): + def rope_encode(self, t, h, w, t_start=0, steps_t=None, steps_h=None, steps_w=None, device=None, dtype=None, transformer_options={}, source_id=0): patch_size = self.patch_size t_len = ((t + (patch_size[0] // 2)) // patch_size[0]) h_len = ((h + (patch_size[1] // 2)) // patch_size[1]) @@ -638,6 +649,13 @@ class WanModel(torch.nn.Module): img_ids = img_ids.reshape(1, -1, img_ids.shape[-1]) freqs = self.rope_embedder(img_ids).movedim(1, 2) + + # In-context reference: a non-zero source_id composes an extra rotation into the spatial rope + if source_id: + d = self.dim // self.num_heads + pos = torch.tensor([[float(source_id)]], device=freqs.device, dtype=torch.float32) + id_rot = rope(pos, d, self.rope_embedder.theta).reshape(1, 1, 1, d // 2, 2, 2).to(freqs.dtype) + freqs = torch.einsum('...ij,...jk->...ik', freqs, id_rot) return freqs def forward(self, x, timestep, context, clip_fea=None, time_dim_concat=None, transformer_options={}, **kwargs): @@ -661,6 +679,15 @@ class WanModel(torch.nn.Module): t_len += 1 freqs = self.rope_encode(t_len, h, w, device=x.device, dtype=x.dtype, transformer_options=transformer_options) + + # In-context reference: one rope block per stream, each with it's own source_id (1, 2, ...) to distinguish from the target (id 0). + context_latents = kwargs.get("context_latents", None) + if context_latents is not None: + context_latents = [comfy.ldm.common_dit.pad_to_patch_size(lat, self.patch_size) for lat in context_latents] + for i, lat in enumerate(context_latents): + freqs = torch.cat([freqs, self.rope_encode(lat.shape[-3], lat.shape[-2], lat.shape[-1], device=x.device, dtype=x.dtype, transformer_options=transformer_options, source_id=i + 1)], dim=1) + kwargs = {**kwargs, "context_latents": context_latents} + return self.forward_orig(x, timestep, context, clip_fea=clip_fea, freqs=freqs, transformer_options=transformer_options, **kwargs)[:, :, :t, :h, :w] def unpatchify(self, x, grid_sizes): diff --git a/comfy/model_base.py b/comfy/model_base.py index d212a7c2a..2a46d1fc1 100644 --- a/comfy/model_base.py +++ b/comfy/model_base.py @@ -1518,8 +1518,26 @@ class WAN21(BaseModel): if reference_latents is not None: out['reference_latent'] = comfy.conds.CONDRegular(self.process_latent_in(reference_latents[-1])[:, :, 0]) + # In-context reference conditioning (Bernini) + context_latents = kwargs.get("context_latents", None) + if context_latents is not None: + out['context_latents'] = comfy.conds.CONDList([self.process_latent_in(l) for l in context_latents]) + return out + def resize_cond_for_context_window(self, cond_key, cond_value, window, x_in, device, retain_index_list=[]): + # In-context cond slicing (Bernini) + if cond_key == "context_latents" and isinstance(getattr(cond_value, "cond", None), list): + dim = window.dim + out = [] + for lat in cond_value.cond: + if lat.ndim > dim and lat.shape[dim] > 1 and lat.shape[dim] == x_in.shape[dim]: + out.append(window.get_tensor(lat, device, dim=dim, retain_index_list=retain_index_list)) + else: + out.append(lat.to(device)) + return cond_value._copy_with(out) + return super().resize_cond_for_context_window(cond_key, cond_value, window, x_in, device, retain_index_list=retain_index_list) + class WAN21_CausalAR(WAN21): def __init__(self, model_config, model_type=ModelType.FLOW, device=None): diff --git a/comfy_extras/nodes_bernini.py b/comfy_extras/nodes_bernini.py new file mode 100644 index 000000000..227fa5753 --- /dev/null +++ b/comfy_extras/nodes_bernini.py @@ -0,0 +1,115 @@ +import torch +from typing_extensions import override + +import comfy.model_management +import comfy.utils +import node_helpers +from comfy_api.latest import ComfyExtension, io + + +def _resize_long_edge(image, max_size, stride=16): + """Resize (preserve aspect) so the long edge <= max_size, then snap each side to `stride`""" + h, w = image.shape[1], image.shape[2] + scale = min(max_size / max(h, w), 1.0) + nh = max(stride, round(h * scale / stride) * stride) + nw = max(stride, round(w * scale / stride) * stride) + return comfy.utils.common_upscale(image[:, :, :, :3].movedim(-1, 1), nw, nh, "area", "disabled").movedim(1, -1) + + +class BerniniConditioning(io.ComfyNode): + """Bernini in-context conditioning for a Wan2.2-A14B model. + + Attaches the VAE-encoded source video / reference images to the conditioning + source video first, then each reference image + + The task is inferred from which inputs are connected: + (nothing) -> t2v (text-to-video) + source_video -> v2v (video-to-video) + source_video + ref_images -> rv2v (reference-guided video editing) + ref_images only -> r2v (reference-to-video) + source_video + ref_video -> ads2v (insert image/video into video) + + source_video is the edit base / canvas (resized to width x height). + reference_video is moving content to composite in. + Streams are ordered source_video, reference_video, then reference_images -> source_id (1, 2, 3, ...). + """ + + @classmethod + def define_schema(cls): + return io.Schema( + node_id="BerniniConditioning", + display_name="Bernini Conditioning", + category="conditioning/video_models", + description="Conditioning node for Bernini in-context video/image conditioning. It can be used for the following tasks: t2v (text-to-video), v2v (video-to-video), rv2v (reference-guided video editing), r2v (reference-to-video), ads2v (insert image/video into video)." + "Reference images injected as in-context tokens (r2v, rv2v) are encoded independently at their own native aspect ratio (long edge capped at ref_max_size)", + inputs=[ + io.Conditioning.Input("positive"), + io.Conditioning.Input("negative"), + io.Vae.Input("vae"), + io.Int.Input("width", default=832, min=16, max=8192, step=16), + io.Int.Input("height", default=480, min=16, max=8192, step=16), + io.Int.Input("length", default=81, min=1, max=8192, step=4), + io.Int.Input("batch_size", default=1, min=1, max=4096), + io.Image.Input("source_video", optional=True, tooltip=( + "Source video to edit or restyle (v2v, rv2v). Resized to width/height and trimmed to length.")), + io.Image.Input("reference_video", optional=True, tooltip=( + "Video to insert into the source video (ads2v).")), + io.Autogrow.Input("reference_images", optional=True, + template=io.Autogrow.TemplatePrefix( + input=io.Image.Input("reference_image", tooltip=( + "Reference image injected as an in-context token (r2v, rv2v).")), + prefix="reference_image_", min=0, max=8)), + io.Int.Input("ref_max_size", default=848, min=16, max=8192, step=16, optional=True, tooltip=( + "Max size for the long edge of reference_video and reference_images. Resized with preserved aspect ratio and snapped to 16px.")), + ], + outputs=[ + io.Conditioning.Output(display_name="positive"), + io.Conditioning.Output(display_name="negative"), + io.Latent.Output(display_name="latent"), + ], + ) + + @classmethod + def execute(cls, positive, negative, vae, width, height, length, batch_size, + source_video=None, reference_video=None, reference_images=None, ref_max_size=848) -> io.NodeOutput: + latent = torch.zeros([batch_size, 16, ((length - 1) // 4) + 1, height // 8, width // 8], + device=comfy.model_management.intermediate_device()) + + # source_video (1), reference_video (2), reference_images (3, 4, ...). + context = [] + if source_video is not None: + vid = comfy.utils.common_upscale(source_video[:length, :, :, :3].movedim(-1, 1), width, height, "area", "center").movedim(1, -1) + context.append(vae.encode(vid[:, :, :, :3])) + + if reference_video is not None: + ref_vid = _resize_long_edge(reference_video[:length], ref_max_size) # moving content, native aspect + context.append(vae.encode(ref_vid[:, :, :, :3])) + + # reference_images is an autogrow dict {reference_image_0: IMAGE, ...}; each slot is a + # separate stream at its own native aspect (a multi-image batch in one slot -> one stream per frame). + if reference_images: + for name in sorted(reference_images): + imgs = reference_images[name] + if imgs is None: + continue + for i in range(imgs.shape[0]): + img = _resize_long_edge(imgs[i:i + 1], ref_max_size) # native aspect per ref + context.append(vae.encode(img[:, :, :, :3])) + + if context: + positive = node_helpers.conditioning_set_values(positive, {"context_latents": context}) + negative = node_helpers.conditioning_set_values(negative, {"context_latents": context}) + + return io.NodeOutput(positive, negative, {"samples": latent}) + + +class BerniniExtension(ComfyExtension): + @override + async def get_node_list(self) -> list[type[io.ComfyNode]]: + return [ + BerniniConditioning, + ] + + +async def comfy_entrypoint() -> BerniniExtension: + return BerniniExtension() diff --git a/nodes.py b/nodes.py index 4bf768045..fb6952bad 100644 --- a/nodes.py +++ b/nodes.py @@ -2404,6 +2404,7 @@ async def init_builtin_extra_nodes(): "nodes_video.py", "nodes_lumina2.py", "nodes_wan.py", + "nodes_bernini.py", "nodes_lotus.py", "nodes_hunyuan3d.py", "nodes_primitive.py", From 5ece24e73c0785e95a45a45a2773ecd1a6f5339b Mon Sep 17 00:00:00 2001 From: Talmaj Date: Wed, 10 Jun 2026 03:28:24 +0200 Subject: [PATCH 04/21] Depth anything 3 (Core-135) (#13853) Co-authored-by: Alexis Rolland --- comfy/image_encoders/dino2.py | 333 ++++++++- comfy/ldm/colormap.py | 25 + comfy/ldm/depth_anything_3/camera.py | 177 +++++ comfy/ldm/depth_anything_3/dpt.py | 489 +++++++++++++ comfy/ldm/depth_anything_3/model.py | 236 ++++++ comfy/ldm/depth_anything_3/preprocess.py | 128 ++++ comfy/ldm/depth_anything_3/ray_pose.py | 272 +++++++ .../reference_view_selector.py | 87 +++ comfy/ldm/depth_anything_3/transform.py | 160 ++++ comfy/model_base.py | 7 + comfy/model_detection.py | 89 +++ comfy/supported_models.py | 18 + comfy_extras/nodes_depth_anything_3.py | 681 ++++++++++++++++++ comfy_extras/nodes_moge.py | 14 +- nodes.py | 3 +- 15 files changed, 2687 insertions(+), 32 deletions(-) create mode 100644 comfy/ldm/colormap.py create mode 100644 comfy/ldm/depth_anything_3/camera.py create mode 100644 comfy/ldm/depth_anything_3/dpt.py create mode 100644 comfy/ldm/depth_anything_3/model.py create mode 100644 comfy/ldm/depth_anything_3/preprocess.py create mode 100644 comfy/ldm/depth_anything_3/ray_pose.py create mode 100644 comfy/ldm/depth_anything_3/reference_view_selector.py create mode 100644 comfy/ldm/depth_anything_3/transform.py create mode 100644 comfy_extras/nodes_depth_anything_3.py diff --git a/comfy/image_encoders/dino2.py b/comfy/image_encoders/dino2.py index ee86f8309..53e4fdb6c 100644 --- a/comfy/image_encoders/dino2.py +++ b/comfy/image_encoders/dino2.py @@ -1,7 +1,13 @@ import torch +import torch.nn.functional as F + from comfy.text_encoders.bert import BertAttention import comfy.model_management from comfy.ldm.modules.attention import optimized_attention_for_device +from comfy.ldm.depth_anything_3.reference_view_selector import ( + select_reference_view, reorder_by_reference, restore_original_order, + THRESH_FOR_REF_SELECTION, +) class Dino2AttentionOutput(torch.nn.Module): @@ -14,13 +20,41 @@ class Dino2AttentionOutput(torch.nn.Module): class Dino2AttentionBlock(torch.nn.Module): - def __init__(self, embed_dim, heads, layer_norm_eps, dtype, device, operations): + def __init__(self, embed_dim, heads, layer_norm_eps, dtype, device, operations, + qk_norm=False): super().__init__() + self.heads = heads + self.head_dim = embed_dim // heads self.attention = BertAttention(embed_dim, heads, dtype, device, operations) self.output = Dino2AttentionOutput(embed_dim, embed_dim, layer_norm_eps, dtype, device, operations) + if qk_norm: + self.q_norm = operations.LayerNorm(self.head_dim, dtype=dtype, device=device) + self.k_norm = operations.LayerNorm(self.head_dim, dtype=dtype, device=device) + else: + self.q_norm = None + self.k_norm = None - def forward(self, x, mask, optimized_attention): - return self.output(self.attention(x, mask, optimized_attention)) + def forward(self, x, mask, optimized_attention, pos=None, rope=None): + # Fast path used by the existing CLIP-vision DINOv2 (no DA3 extensions). + if self.q_norm is None and rope is None: + return self.output(self.attention(x, mask, optimized_attention)) + + # DA3 path: do QKV manually so we can apply per-head QK-norm and 2D RoPE. + attn = self.attention + B, N, C = x.shape + h = self.heads + d = self.head_dim + q = attn.query(x).view(B, N, h, d).transpose(1, 2) + k = attn.key(x).view(B, N, h, d).transpose(1, 2) + v = attn.value(x).view(B, N, h, d).transpose(1, 2) + if self.q_norm is not None: + q = self.q_norm(q) + k = self.k_norm(k) + if rope is not None and pos is not None: + q = rope(q, pos) + k = rope(k, pos) + out = optimized_attention(q, k, v, h, mask=mask, skip_reshape=True) + return self.output(out) class LayerScale(torch.nn.Module): @@ -64,9 +98,11 @@ class SwiGLUFFN(torch.nn.Module): class Dino2Block(torch.nn.Module): - def __init__(self, dim, num_heads, layer_norm_eps, dtype, device, operations, use_swiglu_ffn): + def __init__(self, dim, num_heads, layer_norm_eps, dtype, device, operations, use_swiglu_ffn, + qk_norm=False): super().__init__() - self.attention = Dino2AttentionBlock(dim, num_heads, layer_norm_eps, dtype, device, operations) + self.attention = Dino2AttentionBlock(dim, num_heads, layer_norm_eps, dtype, device, operations, + qk_norm=qk_norm) self.layer_scale1 = LayerScale(dim, dtype, device, operations) self.layer_scale2 = LayerScale(dim, dtype, device, operations) if use_swiglu_ffn: @@ -76,19 +112,90 @@ class Dino2Block(torch.nn.Module): self.norm1 = operations.LayerNorm(dim, eps=layer_norm_eps, dtype=dtype, device=device) self.norm2 = operations.LayerNorm(dim, eps=layer_norm_eps, dtype=dtype, device=device) - def forward(self, x, optimized_attention): - x = x + self.layer_scale1(self.attention(self.norm1(x), None, optimized_attention)) + def forward(self, x, optimized_attention, pos=None, rope=None, attn_mask=None): + x = x + self.layer_scale1(self.attention(self.norm1(x), attn_mask, optimized_attention, + pos=pos, rope=rope)) x = x + self.layer_scale2(self.mlp(self.norm2(x))) return x -class Dino2Encoder(torch.nn.Module): - def __init__(self, dim, num_heads, layer_norm_eps, num_layers, dtype, device, operations, use_swiglu_ffn): +# ----------------------------------------------------------------------------- +# 2D Rotary position embedding (DA3 extension) +# ----------------------------------------------------------------------------- + + +class _PositionGetter: + """Cache (h, w) -> flat (y, x) position grid used to feed ``rope``.""" + + def __init__(self): + self._cache: dict = {} + + def __call__(self, batch_size: int, height: int, width: int, device) -> torch.Tensor: + key = (height, width, device) + if key not in self._cache: + y = torch.arange(height, device=device) + x = torch.arange(width, device=device) + self._cache[key] = torch.cartesian_prod(y, x) + cached = self._cache[key] + return cached.view(1, height * width, 2).expand(batch_size, -1, -1).clone() + + +class RotaryPositionEmbedding2D(torch.nn.Module): + """2D RoPE used by DA3-Small/Base. No learnable parameters.""" + + def __init__(self, frequency: float = 100.0): super().__init__() - self.layer = torch.nn.ModuleList([Dino2Block(dim, num_heads, layer_norm_eps, dtype, device, operations, use_swiglu_ffn = use_swiglu_ffn) - for _ in range(num_layers)]) + self.base_frequency = frequency + self._freq_cache: dict = {} + + def _components(self, dim: int, seq_len: int, device, dtype): + key = (dim, seq_len, device, dtype) + if key not in self._freq_cache: + exp = torch.arange(0, dim, 2, device=device).float() / dim + inv_freq = 1.0 / (self.base_frequency ** exp) + pos = torch.arange(seq_len, device=device, dtype=inv_freq.dtype) + ang = torch.einsum("i,j->ij", pos, inv_freq) + ang = ang.to(dtype) + ang = torch.cat((ang, ang), dim=-1) + self._freq_cache[key] = (ang.cos().to(dtype), ang.sin().to(dtype)) + return self._freq_cache[key] + + @staticmethod + def _rotate(x: torch.Tensor) -> torch.Tensor: + d = x.shape[-1] + x1, x2 = x[..., : d // 2], x[..., d // 2:] + return torch.cat((-x2, x1), dim=-1) + + def _apply_1d(self, tokens, positions, cos_c, sin_c): + cos = F.embedding(positions, cos_c)[:, None, :, :] + sin = F.embedding(positions, sin_c)[:, None, :, :] + return (tokens * cos) + (self._rotate(tokens) * sin) + + def forward(self, tokens: torch.Tensor, positions: torch.Tensor) -> torch.Tensor: + feature_dim = tokens.size(-1) // 2 + max_pos = int(positions.max()) + 1 + cos_c, sin_c = self._components(feature_dim, max_pos, tokens.device, tokens.dtype) + v, h = tokens.chunk(2, dim=-1) + v = self._apply_1d(v, positions[..., 0], cos_c, sin_c) + h = self._apply_1d(h, positions[..., 1], cos_c, sin_c) + return torch.cat((v, h), dim=-1) + + +class Dino2Encoder(torch.nn.Module): + def __init__(self, dim, num_heads, layer_norm_eps, num_layers, dtype, device, operations, use_swiglu_ffn, + qknorm_start: int = -1): + super().__init__() + self.layer = torch.nn.ModuleList([ + Dino2Block( + dim, num_heads, layer_norm_eps, dtype, device, operations, + use_swiglu_ffn=use_swiglu_ffn, + qk_norm=(qknorm_start != -1 and i >= qknorm_start), + ) + for i in range(num_layers) + ]) def forward(self, x, intermediate_output=None): + # Backward-compat path used by ``ClipVisionModel`` (no DA3 extensions). optimized_attention = optimized_attention_for_device(x.device, False, small_input=True) if intermediate_output is not None: @@ -122,16 +229,27 @@ class Dino2PatchEmbeddings(torch.nn.Module): class Dino2Embeddings(torch.nn.Module): - def __init__(self, dim, dtype, device, operations): + def __init__(self, dim, dtype, device, operations, + patch_size: int = 14, image_size: int = 518, + use_mask_token: bool = True, + num_camera_tokens: int = 0): super().__init__() - patch_size = 14 - image_size = 518 self.patch_size = patch_size + self.image_size = image_size self.patch_embeddings = Dino2PatchEmbeddings(dim, patch_size=patch_size, image_size=image_size, dtype=dtype, device=device, operations=operations) self.position_embeddings = torch.nn.Parameter(torch.empty(1, (image_size // patch_size) ** 2 + 1, dim, dtype=dtype, device=device)) self.cls_token = torch.nn.Parameter(torch.empty(1, 1, dim, dtype=dtype, device=device)) # mask_token is a pre-training param, kept only so strict loading accepts the key. - self.mask_token = torch.nn.Parameter(torch.empty(1, dim, dtype=dtype, device=device)) + if use_mask_token: + self.mask_token = torch.nn.Parameter(torch.empty(1, dim, dtype=dtype, device=device)) + else: + self.mask_token = None + if num_camera_tokens > 0: + # DA3 stores (ref_token, src_token) pairs that get injected at the + # alt-attn boundary; see ``Dinov2Model._inject_camera_token``. + self.camera_token = torch.nn.Parameter(torch.empty(1, num_camera_tokens, dim, dtype=dtype, device=device)) + else: + self.camera_token = None def interpolate_pos_encoding(self, x, h_pixels, w_pixels): pos_embed = comfy.model_management.cast_to_device(self.position_embeddings, x.device, torch.float32) @@ -140,12 +258,22 @@ class Dino2Embeddings(torch.nn.Module): patch_pos = pos_embed[:, 1:] N = patch_pos.shape[1] M = int(N ** 0.5) + assert N == M * M, f"DINOv2 position grid must be square, got N={N} patches (sqrt={M})" h0 = h_pixels // self.patch_size w0 = w_pixels // self.patch_size - scale_factor = ((h0 + 0.1) / M, (w0 + 0.1) / M) # +0.1 matches upstream DINOv2's FP-rounding workaround so the interpolate output size lands on (h0, w0). + # +0.1 matches upstream DINOv2's FP-rounding workaround so the interpolate output size lands on (h0, w0). + # scale_factor is (height_scale, width_scale) -- height MUST come first; + # swapping these only happens to work for square inputs and breaks + # non-square paths like DA3-Small / DA3-Base multi-view. + scale_factor = ((h0 + 0.1) / M, (w0 + 0.1) / M) patch_pos = patch_pos.reshape(1, M, M, -1).permute(0, 3, 1, 2) patch_pos = torch.nn.functional.interpolate(patch_pos, scale_factor=scale_factor, mode="bicubic", antialias=False) + assert (h0, w0) == patch_pos.shape[-2:], ( + f"Interpolated pos-embed grid {tuple(patch_pos.shape[-2:])} does not match " + f"target patch grid ({h0}, {w0}) for input {h_pixels}x{w_pixels} (patch_size={self.patch_size}); " + f"check scale_factor axis order and +0.1 rounding workaround" + ) patch_pos = patch_pos.permute(0, 2, 3, 1).flatten(1, 2) return torch.cat((class_pos, patch_pos), dim=1).to(x.dtype) @@ -168,12 +296,51 @@ class Dinov2Model(torch.nn.Module): heads = config_dict["num_attention_heads"] layer_norm_eps = config_dict["layer_norm_eps"] use_swiglu_ffn = config_dict["use_swiglu_ffn"] + patch_size = config_dict.get("patch_size", 14) + image_size = config_dict.get("image_size", 518) + use_mask_token = config_dict.get("use_mask_token", True) - self.embeddings = Dino2Embeddings(dim, dtype, device, operations) - self.encoder = Dino2Encoder(dim, heads, layer_norm_eps, num_layers, dtype, device, operations, use_swiglu_ffn = use_swiglu_ffn) + # DA3 extensions (all default to disabled). + self.alt_start = config_dict.get("alt_start", -1) + self.qknorm_start = config_dict.get("qknorm_start", -1) + self.rope_start = config_dict.get("rope_start", -1) + self.cat_token = config_dict.get("cat_token", False) + rope_freq = config_dict.get("rope_freq", 100.0) + + self.embed_dim = dim + self.patch_size = patch_size + self.num_register_tokens = 0 + self.patch_start_idx = 1 + + if self.rope_start != -1 and rope_freq > 0: + self.rope = RotaryPositionEmbedding2D(frequency=rope_freq) + self._position_getter = _PositionGetter() + else: + self.rope = None + self._position_getter = None + + # camera_token shape: (1, 2, dim) -> (ref_token, src_token). + num_cam_tokens = 2 if self.alt_start != -1 else 0 + + self.embeddings = Dino2Embeddings( + dim, dtype, device, operations, + patch_size=patch_size, image_size=image_size, + use_mask_token=use_mask_token, num_camera_tokens=num_cam_tokens, + ) + self.encoder = Dino2Encoder( + dim, heads, layer_norm_eps, num_layers, dtype, device, operations, + use_swiglu_ffn=use_swiglu_ffn, + qknorm_start=self.qknorm_start, + ) self.layernorm = operations.LayerNorm(dim, eps=layer_norm_eps, dtype=dtype, device=device) def forward(self, pixel_values, attention_mask=None, intermediate_output=None): + if self.alt_start != -1: + raise RuntimeError( + "Dinov2Model.forward() is the backward-compatible CLIP-vision path and does not " + "apply DA3 extensions (RoPE, alternating attention, camera-token injection). " + "Use get_intermediate_layers_da3() for Depth Anything 3 models." + ) x = self.embeddings(pixel_values) x, i = self.encoder(x, intermediate_output=intermediate_output) x = self.layernorm(x) @@ -181,6 +348,7 @@ class Dinov2Model(torch.nn.Module): return x, i, pooled_output, None def get_intermediate_layers(self, pixel_values, indices, apply_norm=True): + """Single-view multi-layer feature extraction.""" x = self.embeddings(pixel_values) optimized_attention = optimized_attention_for_device(x.device, False, small_input=True) n_layers = len(self.encoder.layer) @@ -197,3 +365,132 @@ class Dinov2Model(torch.nn.Module): if i >= max_idx: break return [cache[i] for i in resolved] + + # ------------------------------------------------------------------ + # Depth Anything 3 forward + # ------------------------------------------------------------------ + def _prepare_rope_positions(self, B, S, H, W, device): + if self.rope is None: + return None, None + ph, pw = H // self.patch_size, W // self.patch_size + pos = self._position_getter(B * S, ph, pw, device=device) + # Shift so the cls/cam token at position 0 is reserved for "no diff". + pos = pos + 1 + cls_pos = torch.zeros(B * S, self.patch_start_idx, 2, device=device, dtype=pos.dtype) + # Per-view local: real grid positions for patches, 0 for cls token. + pos_local = torch.cat([cls_pos, pos], dim=1) + # Global (across views): same grid positions; cls token still at 0, + # but patches share the same positions in every view. + pos_global = torch.cat([cls_pos, torch.zeros_like(pos) + 1], dim=1) + return pos_local, pos_global + + def _inject_camera_token(self, x: torch.Tensor, B: int, S: int, cam_token: "torch.Tensor | None") -> torch.Tensor: + # x: (B, S, N, C). Replace token at index 0 with the camera token. + if cam_token is not None: + inj = cam_token + else: + ct = comfy.model_management.cast_to_device(self.embeddings.camera_token, x.device, x.dtype) + ref_token = ct[:, :1].expand(B, -1, -1) + src_token = ct[:, 1:].expand(B, max(S - 1, 0), -1) + inj = torch.cat([ref_token, src_token], dim=1) + x = x.clone() + x[:, :, 0] = inj + return x + + def get_intermediate_layers_da3(self, pixel_values, out_layers, cam_token=None, ref_view_strategy="saddle_balanced", export_feat_layers=None): + """Multi-view multi-layer feature extraction used by Depth Anything 3.""" + if pixel_values.ndim == 4: + pixel_values = pixel_values.unsqueeze(1) + assert pixel_values.ndim == 5 and pixel_values.shape[2] == 3, \ + f"expected (B,3,H,W) or (B,S,3,H,W); got {tuple(pixel_values.shape)}" + B, S, _, H, W = pixel_values.shape + + # Patch + cls + (interpolated) pos embed for each view. + x = pixel_values.reshape(B * S, 3, H, W) + x = self.embeddings(x) # (B*S, 1+N, C) + x = x.reshape(B, S, x.shape[-2], x.shape[-1]) # (B, S, 1+N, C) + + pos_local, pos_global = self._prepare_rope_positions(B, S, H, W, x.device) + # optimized_attention is only used by blocks without QK-norm/RoPE + # (vanilla DINOv2 path); enabling-aware blocks fall through to SDPA. + optimized_attention = optimized_attention_for_device(x.device, False, small_input=True) + + out_set = set(out_layers) + export_set = set(export_feat_layers) if export_feat_layers else set() + outputs: list[torch.Tensor] = [] + aux_outputs: list[torch.Tensor] = [] + local_x = x + b_idx = None + + + for i, blk in enumerate(self.encoder.layer): + apply_rope = self.rope is not None and i >= self.rope_start + block_rope = self.rope if apply_rope else None + l_pos = pos_local if apply_rope else None + g_pos = pos_global if apply_rope else None + + # Reference-view selection threshold: matches the upstream constant + # THRESH_FOR_REF_SELECTION = 3. Skipped when a user-supplied + # cam_token is provided (camera info already pins the geometry). + if (self.alt_start != -1 and i == self.alt_start - 1 and S >= THRESH_FOR_REF_SELECTION and cam_token is None): + b_idx = select_reference_view(x, strategy=ref_view_strategy) + x = reorder_by_reference(x, b_idx) + local_x = reorder_by_reference(local_x, b_idx) + + if self.alt_start != -1 and i == self.alt_start: + x = self._inject_camera_token(x, B, S, cam_token) + + if self.alt_start != -1 and i >= self.alt_start and (i % 2 == 1): + # Global attention across views: flatten S into the seq dim. + t = x.reshape(B, S * x.shape[-2], x.shape[-1]) + p = g_pos.reshape(B, S * g_pos.shape[-2], g_pos.shape[-1]) if g_pos is not None else None + t = blk(t, optimized_attention=optimized_attention, pos=p, rope=block_rope) + x = t.reshape(B, S, x.shape[-2], x.shape[-1]) + else: + # Per-view local attention. + t = x.reshape(B * S, x.shape[-2], x.shape[-1]) + p = l_pos.reshape(B * S, l_pos.shape[-2], l_pos.shape[-1]) if l_pos is not None else None + t = blk(t, optimized_attention=optimized_attention, pos=p, rope=block_rope) + x = t.reshape(B, S, x.shape[-2], x.shape[-1]) + local_x = x + + if i in out_set: + if self.cat_token: + out_x = torch.cat([local_x, x], dim=-1) + else: + out_x = x + # Restore original view order on the way out so heads see views + # in the user's expected order. + if b_idx is not None and self.alt_start != -1: + out_x = restore_original_order(out_x, b_idx) + outputs.append(out_x) + + if i in export_set: + aux = x + if b_idx is not None and self.alt_start != -1: + aux = restore_original_order(aux, b_idx) + aux_outputs.append(aux) + + # Apply final norm. When cat_token is set, only the right half + # ("global" features) is normalised; the left half is left as-is to + # match the upstream DA3 head signature. + normed: list[torch.Tensor] = [] + cls_tokens: list[torch.Tensor] = [] + for out_x in outputs: + cls_tokens.append(out_x[:, :, 0]) + if out_x.shape[-1] == self.embed_dim: + normed.append(self.layernorm(out_x)) + elif out_x.shape[-1] == self.embed_dim * 2: + left = out_x[..., :self.embed_dim] + right = self.layernorm(out_x[..., self.embed_dim:]) + normed.append(torch.cat([left, right], dim=-1)) + else: + raise ValueError(f"Unexpected token width: {out_x.shape[-1]}") + + # Drop cls/cam token from the patch sequence. + normed = [o[..., 1 + self.num_register_tokens:, :] for o in normed] + + # Final layernorm + drop cls token from auxiliary features too. + aux_normed = [self.layernorm(o)[..., 1 + self.num_register_tokens:, :] + for o in aux_outputs] + return list(zip(normed, cls_tokens)), aux_normed diff --git a/comfy/ldm/colormap.py b/comfy/ldm/colormap.py new file mode 100644 index 000000000..1f4d88bd9 --- /dev/null +++ b/comfy/ldm/colormap.py @@ -0,0 +1,25 @@ +"""Colormap utilities for depth and geometry visualisation.""" + +from __future__ import annotations + +import torch + + +def turbo(x: torch.Tensor) -> torch.Tensor: + """Anton Mikhailov polynomial approximation of the Turbo colormap. + + Args: + x: Float tensor with values in [0, 1]. + + Returns: + RGB tensor of the same shape as ``x`` with a trailing size-3 dimension. + """ + x = x.clamp(0.0, 1.0) + x2 = x * x + x3 = x2 * x + x4 = x2 * x2 + x5 = x4 * x + r = 0.13572138 + 4.61539260*x - 42.66032258*x2 + 132.13108234*x3 - 152.94239396*x4 + 59.28637943*x5 + g = 0.09140261 + 2.19418839*x + 4.84296658*x2 - 14.18503333*x3 + 4.27729857*x4 + 2.82956604*x5 + b = 0.10667330 + 12.64194608*x - 60.58204836*x2 + 110.36276771*x3 - 89.90310912*x4 + 27.34824973*x5 + return torch.stack([r, g, b], dim=-1).clamp(0.0, 1.0) diff --git a/comfy/ldm/depth_anything_3/camera.py b/comfy/ldm/depth_anything_3/camera.py new file mode 100644 index 000000000..65a57d66f --- /dev/null +++ b/comfy/ldm/depth_anything_3/camera.py @@ -0,0 +1,177 @@ +"""Camera-token encoder and decoder for Depth Anything 3.""" + +from __future__ import annotations + +import torch +import torch.nn as nn +import torch.nn.functional as F + +from comfy.ldm.modules.attention import optimized_attention_for_device +from .transform import affine_inverse, extri_intri_to_pose_encoding + + +# ----------------------------------------------------------------------- +# Building blocks (mirror depth_anything_3.model.utils.{attention,block}) +# ----------------------------------------------------------------------- + + +class _Mlp(nn.Module): + """Standard 2-layer MLP with GELU. Matches upstream ``utils.attention.Mlp``.""" + + def __init__(self, in_features, hidden_features=None, out_features=None, *, device=None, dtype=None, operations=None): + super().__init__() + out_features = out_features or in_features + hidden_features = hidden_features or in_features + self.fc1 = operations.Linear(in_features, hidden_features, bias=True, device=device, dtype=dtype) + self.fc2 = operations.Linear(hidden_features, out_features, bias=True, device=device, dtype=dtype) + + def forward(self, x): + return self.fc2(F.gelu(self.fc1(x))) + + +class _LayerScale(nn.Module): + """Per-channel learnable scaling. Matches upstream LayerScale.""" + + def __init__(self, dim, *, device=None, dtype=None): + super().__init__() + self.gamma = nn.Parameter(torch.empty(dim, device=device, dtype=dtype)) + + def forward(self, x): + return x * self.gamma.to(dtype=x.dtype, device=x.device) + + +class _Attention(nn.Module): + """ Self-attention with fused QKV projection. Mirrors upstream utils.attention.Attention; + Layout matches the HF safetensors (attn.qkv.{weight,bias} and attn.proj.{weight,bias}).""" + + def __init__(self, dim, num_heads, *, device=None, dtype=None, operations=None): + super().__init__() + assert dim % num_heads == 0 + self.num_heads = num_heads + self.head_dim = dim // num_heads + self.qkv = operations.Linear(dim, dim * 3, bias=True, device=device, dtype=dtype) + self.proj = operations.Linear(dim, dim, bias=True, device=device, dtype=dtype) + + def forward(self, x): + B, N, C = x.shape + qkv = self.qkv(x).reshape(B, N, 3, C) + q, k, v = qkv.unbind(2) # each (B, N, C) + attn_fn = optimized_attention_for_device(x.device, small_input=True) + out = attn_fn(q, k, v, heads=self.num_heads) + return self.proj(out) + + +class _Block(nn.Module): + """Pre-norm transformer block with LayerScale. Used by :class:CameraEnc. Layout follows upstream utils.block.Block.""" + + def __init__(self, dim, num_heads, mlp_ratio=4, init_values=0.01, *, device=None, dtype=None, operations=None): + super().__init__() + self.norm1 = operations.LayerNorm(dim, device=device, dtype=dtype) + self.attn = _Attention(dim, num_heads, device=device, dtype=dtype, operations=operations) + self.ls1 = _LayerScale(dim, device=device, dtype=dtype) if init_values else nn.Identity() + self.norm2 = operations.LayerNorm(dim, device=device, dtype=dtype) + self.mlp = _Mlp(in_features=dim, hidden_features=int(dim * mlp_ratio), device=device, dtype=dtype, operations=operations) + self.ls2 = _LayerScale(dim, device=device, dtype=dtype) if init_values else nn.Identity() + + def forward(self, x): + x = x + self.ls1(self.attn(self.norm1(x))) + x = x + self.ls2(self.mlp(self.norm2(x))) + return x + + +class CameraEnc(nn.Module): + """Encode per-view (extrinsics, intrinsics) into a camera token. + + Maps a 9-D pose-encoding vector through a small MLP up to the backbone's + ``embed_dim``, then runs ``trunk_depth`` transformer blocks. The output + has shape ``(B, S, embed_dim)`` and is injected at block ``alt_start`` + of the DINOv2 backbone in place of the cls token. + + Parameters mirror the upstream ``cam_enc.py`` so HF weights load directly. + """ + + def __init__( + self, + dim_out: int = 1024, + dim_in: int = 9, + trunk_depth: int = 4, + target_dim: int = 9, + num_heads: int = 16, + mlp_ratio: int = 4, + init_values: float = 0.01, + *, + device=None, dtype=None, operations=None, + **_kwargs, + ): + super().__init__() + self.target_dim = target_dim + self.trunk_depth = trunk_depth + self.trunk = nn.Sequential(*[ + _Block(dim_out, num_heads=num_heads, mlp_ratio=mlp_ratio, + init_values=init_values, + device=device, dtype=dtype, operations=operations) + for _ in range(trunk_depth) + ]) + self.token_norm = operations.LayerNorm(dim_out, device=device, dtype=dtype) + self.trunk_norm = operations.LayerNorm(dim_out, device=device, dtype=dtype) + self.pose_branch = _Mlp( + in_features=dim_in, + hidden_features=dim_out // 2, + out_features=dim_out, + device=device, dtype=dtype, operations=operations, + ) + + def forward(self, extrinsics: torch.Tensor, intrinsics: torch.Tensor, + image_size_hw) -> torch.Tensor: + """Encode camera parameters into ``(B, S, dim_out)`` tokens.""" + c2ws = affine_inverse(extrinsics) + pose_encoding = extri_intri_to_pose_encoding(c2ws, intrinsics, image_size_hw) + tokens = self.pose_branch(pose_encoding.to(self.pose_branch.fc1.weight.dtype)) + tokens = self.token_norm(tokens) + tokens = self.trunk(tokens) + tokens = self.trunk_norm(tokens) + return tokens + + +class CameraDec(nn.Module): + """Decode the final cam token into a 9-D pose encoding. + + Output layout: ``[T(3), quat_xyzw(4), fov_h, fov_w]``. The translation is + always predicted by the network; the quaternion and FoV can either be + predicted or supplied via ``camera_encoding`` (used at training time + when GT cameras are available -- not exercised at inference here). + + Parameters mirror the upstream ``cam_dec.py`` so HF weights load directly. + """ + + def __init__(self, dim_in: int = 1536, + *, device=None, dtype=None, operations=None, **_kwargs): + super().__init__() + d = dim_in + self.backbone = nn.Sequential( + operations.Linear(d, d, device=device, dtype=dtype), + nn.ReLU(), + operations.Linear(d, d, device=device, dtype=dtype), + nn.ReLU(), + ) + self.fc_t = operations.Linear(d, 3, device=device, dtype=dtype) + self.fc_qvec = operations.Linear(d, 4, device=device, dtype=dtype) + self.fc_fov = nn.Sequential( + operations.Linear(d, 2, device=device, dtype=dtype), + nn.ReLU(), + ) + + def forward(self, feat: torch.Tensor, + camera_encoding: "torch.Tensor | None" = None) -> torch.Tensor: + """Decode ``(B, N, dim_in)`` cam tokens into ``(B, N, 9)`` pose enc.""" + B, N = feat.shape[:2] + feat = feat.reshape(B * N, -1) + feat = self.backbone(feat) + out_t = self.fc_t(feat.float()).reshape(B, N, 3) + if camera_encoding is None: + out_qvec = self.fc_qvec(feat.float()).reshape(B, N, 4) + out_fov = self.fc_fov(feat.float()).reshape(B, N, 2) + else: + out_qvec = camera_encoding[..., 3:7] + out_fov = camera_encoding[..., -2:] + return torch.cat([out_t, out_qvec, out_fov], dim=-1) diff --git a/comfy/ldm/depth_anything_3/dpt.py b/comfy/ldm/depth_anything_3/dpt.py new file mode 100644 index 000000000..fb940873b --- /dev/null +++ b/comfy/ldm/depth_anything_3/dpt.py @@ -0,0 +1,489 @@ +"""DPT / DualDPT heads for Depth Anything 3.""" + +from __future__ import annotations + +from typing import List, Optional, Sequence, Tuple + +import torch +import torch.nn as nn +import torch.nn.functional as F + + +class Permute(nn.Module): + def __init__(self, dims: Tuple[int, ...]): + super().__init__() + self.dims = dims + + def forward(self, x: torch.Tensor) -> torch.Tensor: + return x.permute(*self.dims) + + +def _custom_interpolate( + x: torch.Tensor, + size: Optional[Tuple[int, int]] = None, + scale_factor: Optional[float] = None, + mode: str = "bilinear", + align_corners: bool = True, +) -> torch.Tensor: + if size is None: + assert scale_factor is not None + size = (int(x.shape[-2] * scale_factor), int(x.shape[-1] * scale_factor)) + INT_MAX = 1610612736 + total = size[0] * size[1] * x.shape[0] * x.shape[1] + if total > INT_MAX: + chunks = torch.chunk(x, chunks=(total // INT_MAX) + 1, dim=0) + outs = [F.interpolate(c, size=size, mode=mode, align_corners=align_corners) for c in chunks] + return torch.cat(outs, dim=0).contiguous() + return F.interpolate(x, size=size, mode=mode, align_corners=align_corners) + + +def _create_uv_grid(width: int, height: int, aspect_ratio: float, dtype, device) -> torch.Tensor: + """Normalised UV grid spanning (-x_span, -y_span)..(x_span, y_span).""" + diag_factor = (aspect_ratio ** 2 + 1.0) ** 0.5 + span_x = aspect_ratio / diag_factor + span_y = 1.0 / diag_factor + left_x = -span_x * (width - 1) / width + right_x = span_x * (width - 1) / width + top_y = -span_y * (height - 1) / height + bottom_y = span_y * (height - 1) / height + x_coords = torch.linspace(left_x, right_x, steps=width, dtype=dtype, device=device) + y_coords = torch.linspace(top_y, bottom_y, steps=height, dtype=dtype, device=device) + uu, vv = torch.meshgrid(x_coords, y_coords, indexing="xy") + return torch.stack((uu, vv), dim=-1) # (H, W, 2) + + +def _make_sincos_pos_embed(embed_dim: int, pos: torch.Tensor, omega_0: float = 100.0) -> torch.Tensor: + omega = torch.arange(embed_dim // 2, dtype=torch.float32, device=pos.device) + omega = 1.0 / omega_0 ** (omega / (embed_dim / 2.0)) + pos = pos.reshape(-1) + out = torch.einsum("m,d->md", pos, omega) + return torch.cat([out.sin(), out.cos()], dim=1).float() + + +def _position_grid_to_embed(pos_grid: torch.Tensor, embed_dim: int, omega_0: float = 100.0) -> torch.Tensor: + H, W, _ = pos_grid.shape + pos_flat = pos_grid.reshape(-1, 2) + emb_x = _make_sincos_pos_embed(embed_dim // 2, pos_flat[:, 0], omega_0=omega_0) + emb_y = _make_sincos_pos_embed(embed_dim // 2, pos_flat[:, 1], omega_0=omega_0) + emb = torch.cat([emb_x, emb_y], dim=-1) + return emb.view(H, W, embed_dim) + + +def _add_pos_embed(x: torch.Tensor, W: int, H: int, ratio: float = 0.1) -> torch.Tensor: + """Stateless UV positional embedding added to a feature map (B, C, h, w).""" + pw, ph = x.shape[-1], x.shape[-2] + pe = _create_uv_grid(pw, ph, aspect_ratio=W / H, dtype=x.dtype, device=x.device) + pe = _position_grid_to_embed(pe, x.shape[1]) * ratio + pe = pe.permute(2, 0, 1)[None].expand(x.shape[0], -1, -1, -1).to(dtype=x.dtype) + return x + pe + + +def _apply_activation(x: torch.Tensor, activation: str) -> torch.Tensor: + act = (activation or "linear").lower() + if act == "exp": + return torch.exp(x) + if act == "expp1": + return torch.exp(x) + 1 + if act == "expm1": + return torch.expm1(x) + if act == "relu": + return torch.relu(x) + if act == "sigmoid": + return torch.sigmoid(x) + if act == "softplus": + return F.softplus(x) + if act == "tanh": + return torch.tanh(x) + return x + + +# ----------------------------------------------------------------------------- +# Fusion building blocks +# ----------------------------------------------------------------------------- + + +class ResidualConvUnit(nn.Module): + def __init__(self, features: int, device=None, dtype=None, operations=None): + super().__init__() + self.conv1 = operations.Conv2d(features, features, 3, 1, 1, bias=True, device=device, dtype=dtype) + self.conv2 = operations.Conv2d(features, features, 3, 1, 1, bias=True, device=device, dtype=dtype) + self.activation = nn.ReLU(inplace=False) + + def forward(self, x: torch.Tensor) -> torch.Tensor: + out = self.activation(x) + out = self.conv1(out) + out = self.activation(out) + out = self.conv2(out) + return out + x + + +class FeatureFusionBlock(nn.Module): + def __init__(self, features: int, has_residual: bool = True, align_corners: bool = True, device=None, dtype=None, operations=None): + super().__init__() + self.align_corners = align_corners + self.has_residual = has_residual + if has_residual: + self.resConfUnit1 = ResidualConvUnit(features, device=device, dtype=dtype, operations=operations) + else: + self.resConfUnit1 = None + self.resConfUnit2 = ResidualConvUnit(features, device=device, dtype=dtype, operations=operations) + self.out_conv = operations.Conv2d(features, features, 1, 1, 0, bias=True, device=device, dtype=dtype) + + def forward(self, *xs: torch.Tensor, size: Optional[Tuple[int, int]] = None) -> torch.Tensor: + y = xs[0] + if self.has_residual and len(xs) > 1 and self.resConfUnit1 is not None: + y = y + self.resConfUnit1(xs[1]) + y = self.resConfUnit2(y) + if size is None: + up_kwargs = {"scale_factor": 2.0} + else: + up_kwargs = {"size": size} + y = _custom_interpolate(y, **up_kwargs, mode="bilinear", align_corners=self.align_corners) + y = self.out_conv(y) + return y + + +class _Scratch(nn.Module): + """Container that mirrors upstream ``scratch`` attribute layout.""" + + +def _make_scratch(in_shape: List[int], out_shape: int, device=None, dtype=None, operations=None) -> _Scratch: + scratch = _Scratch() + scratch.layer1_rn = operations.Conv2d(in_shape[0], out_shape, 3, 1, 1, bias=False, device=device, dtype=dtype) + scratch.layer2_rn = operations.Conv2d(in_shape[1], out_shape, 3, 1, 1, bias=False, device=device, dtype=dtype) + scratch.layer3_rn = operations.Conv2d(in_shape[2], out_shape, 3, 1, 1, bias=False, device=device, dtype=dtype) + scratch.layer4_rn = operations.Conv2d(in_shape[3], out_shape, 3, 1, 1, bias=False, device=device, dtype=dtype) + return scratch + + +def _make_fusion_block(features: int, has_residual: bool = True, device=None, dtype=None, operations=None) -> FeatureFusionBlock: + return FeatureFusionBlock(features, has_residual=has_residual, align_corners=True, device=device, dtype=dtype, operations=operations) + + +# ----------------------------------------------------------------------------- +# DPT (single head + optional sky head) -- used by DA3Mono/Metric +# ----------------------------------------------------------------------------- + + +class DPT(nn.Module): + """Single-head DPT used by DA3Mono-Large and DA3Metric-Large.""" + + def __init__( + self, + dim_in: int, + patch_size: int = 14, + output_dim: int = 1, + activation: str = "exp", + conf_activation: str = "expp1", + features: int = 256, + out_channels: Sequence[int] = (256, 512, 1024, 1024), + pos_embed: bool = False, + down_ratio: int = 1, + head_name: str = "depth", + use_sky_head: bool = True, + sky_name: str = "sky", + sky_activation: str = "relu", + norm_type: str = "idt", + device=None, dtype=None, operations=None, + ): + super().__init__() + self.patch_size = patch_size + self.activation = activation + self.conf_activation = conf_activation + self.pos_embed = pos_embed + self.down_ratio = down_ratio + self.head_main = head_name + self.sky_name = sky_name + self.out_dim = output_dim + self.has_conf = output_dim > 1 + self.use_sky_head = use_sky_head + self.sky_activation = sky_activation + self.intermediate_layer_idx: Tuple[int, int, int, int] = (0, 1, 2, 3) + + if norm_type == "layer": + self.norm = operations.LayerNorm(dim_in, device=device, dtype=dtype) + else: + self.norm = nn.Identity() + + out_channels = list(out_channels) + self.projects = nn.ModuleList([ + operations.Conv2d(dim_in, oc, kernel_size=1, stride=1, padding=0, device=device, dtype=dtype) + for oc in out_channels + ]) + self.resize_layers = nn.ModuleList([ + operations.ConvTranspose2d(out_channels[0], out_channels[0], kernel_size=4, stride=4, padding=0, device=device, dtype=dtype), + operations.ConvTranspose2d(out_channels[1], out_channels[1], kernel_size=2, stride=2, padding=0, device=device, dtype=dtype), + nn.Identity(), + operations.Conv2d(out_channels[3], out_channels[3], kernel_size=3, stride=2, padding=1, device=device, dtype=dtype), + ]) + + self.scratch = _make_scratch(out_channels, features, device=device, dtype=dtype, operations=operations) + self.scratch.refinenet1 = _make_fusion_block(features, device=device, dtype=dtype, operations=operations) + self.scratch.refinenet2 = _make_fusion_block(features, device=device, dtype=dtype, operations=operations) + self.scratch.refinenet3 = _make_fusion_block(features, device=device, dtype=dtype, operations=operations) + self.scratch.refinenet4 = _make_fusion_block(features, has_residual=False, device=device, dtype=dtype, operations=operations) + + head_features_1 = features + head_features_2 = 32 + self.scratch.output_conv1 = operations.Conv2d( + head_features_1, head_features_1 // 2, kernel_size=3, stride=1, padding=1, + device=device, dtype=dtype, + ) + self.scratch.output_conv2 = nn.Sequential( + operations.Conv2d(head_features_1 // 2, head_features_2, kernel_size=3, stride=1, padding=1, device=device, dtype=dtype), + nn.ReLU(inplace=False), + operations.Conv2d(head_features_2, output_dim, kernel_size=1, stride=1, padding=0, device=device, dtype=dtype), + ) + + if self.use_sky_head: + self.scratch.sky_output_conv2 = nn.Sequential( + operations.Conv2d(head_features_1 // 2, head_features_2, kernel_size=3, stride=1, padding=1, device=device, dtype=dtype), + nn.ReLU(inplace=False), + operations.Conv2d(head_features_2, 1, kernel_size=1, stride=1, padding=0, device=device, dtype=dtype), + ) + + def forward(self, feats: List[torch.Tensor], H: int, W: int, patch_start_idx: int = 0, **_kwargs) -> dict: + # feats[i][0] is the patch-token tensor with shape (B, S, N_patch, C) + B, S, N, C = feats[0][0].shape + feats_flat = [feat[0].reshape(B * S, N, C) for feat in feats] + + ph, pw = H // self.patch_size, W // self.patch_size + resized = [] + for stage_idx, take_idx in enumerate(self.intermediate_layer_idx): + x = feats_flat[take_idx][:, patch_start_idx:] + x = self.norm(x) + x = x.permute(0, 2, 1).contiguous().reshape(B * S, C, ph, pw) + x = self.projects[stage_idx](x) + if self.pos_embed: + x = _add_pos_embed(x, W, H) + x = self.resize_layers[stage_idx](x) + resized.append(x) + + l1_rn = self.scratch.layer1_rn(resized[0]) + l2_rn = self.scratch.layer2_rn(resized[1]) + l3_rn = self.scratch.layer3_rn(resized[2]) + l4_rn = self.scratch.layer4_rn(resized[3]) + + out = self.scratch.refinenet4(l4_rn, size=l3_rn.shape[2:]) + out = self.scratch.refinenet3(out, l3_rn, size=l2_rn.shape[2:]) + out = self.scratch.refinenet2(out, l2_rn, size=l1_rn.shape[2:]) + out = self.scratch.refinenet1(out, l1_rn) + + h_out = int(ph * self.patch_size / self.down_ratio) + w_out = int(pw * self.patch_size / self.down_ratio) + + fused = self.scratch.output_conv1(out) + fused = _custom_interpolate(fused, (h_out, w_out), mode="bilinear", align_corners=True) + if self.pos_embed: + fused = _add_pos_embed(fused, W, H) + feat = fused + + main_logits = self.scratch.output_conv2(feat) + outs = {} + if self.has_conf: + fmap = main_logits.permute(0, 2, 3, 1) + pred = _apply_activation(fmap[..., :-1], self.activation) + conf = _apply_activation(fmap[..., -1], self.conf_activation) + outs[self.head_main] = pred.squeeze(-1).view(B, S, *pred.shape[1:-1]) + outs[f"{self.head_main}_conf"] = conf.view(B, S, *conf.shape[1:]) + else: + pred = _apply_activation(main_logits, self.activation) + outs[self.head_main] = pred.squeeze(1).view(B, S, *pred.shape[2:]) + + if self.use_sky_head: + sky_logits = self.scratch.sky_output_conv2(feat) + if self.sky_activation.lower() == "sigmoid": + sky = torch.sigmoid(sky_logits) + elif self.sky_activation.lower() == "relu": + sky = F.relu(sky_logits) + else: + sky = sky_logits + outs[self.sky_name] = sky.squeeze(1).view(B, S, *sky.shape[2:]) + + return outs + + +# ----------------------------------------------------------------------------- +# DualDPT (depth + auxiliary "ray" head) -- used by DA3-Small / DA3-Base +# ----------------------------------------------------------------------------- + + +class DualDPT(nn.Module): + """Two-head DPT used by DA3-Small / DA3-Base.""" + + def __init__( + self, + dim_in: int, + patch_size: int = 14, + output_dim: int = 2, + activation: str = "exp", + conf_activation: str = "expp1", + features: int = 256, + out_channels: Sequence[int] = (256, 512, 1024, 1024), + pos_embed: bool = True, + down_ratio: int = 1, + aux_pyramid_levels: int = 4, + aux_out1_conv_num: int = 5, + head_names: Tuple[str, str] = ("depth", "ray"), + device=None, dtype=None, operations=None, + ): + super().__init__() + self.patch_size = patch_size + self.activation = activation + self.conf_activation = conf_activation + self.pos_embed = pos_embed + self.down_ratio = down_ratio + self.aux_levels = aux_pyramid_levels + self.aux_out1_conv_num = aux_out1_conv_num + self.head_main, self.head_aux = head_names + self.intermediate_layer_idx: Tuple[int, int, int, int] = (0, 1, 2, 3) + # Toggle the auxiliary ray branch at runtime. Default off (mono path). + # DepthAnything3Net flips this on when running multi-view + ray-pose. + self.enable_aux: bool = False + + self.norm = operations.LayerNorm(dim_in, device=device, dtype=dtype) + out_channels = list(out_channels) + self.projects = nn.ModuleList([ + operations.Conv2d(dim_in, oc, kernel_size=1, stride=1, padding=0, device=device, dtype=dtype) + for oc in out_channels + ]) + self.resize_layers = nn.ModuleList([ + operations.ConvTranspose2d(out_channels[0], out_channels[0], kernel_size=4, stride=4, padding=0, device=device, dtype=dtype), + operations.ConvTranspose2d(out_channels[1], out_channels[1], kernel_size=2, stride=2, padding=0, device=device, dtype=dtype), + nn.Identity(), + operations.Conv2d(out_channels[3], out_channels[3], kernel_size=3, stride=2, padding=1, device=device, dtype=dtype), + ]) + + self.scratch = _make_scratch(out_channels, features, device=device, dtype=dtype, operations=operations) + # Main fusion chain + self.scratch.refinenet1 = _make_fusion_block(features, device=device, dtype=dtype, operations=operations) + self.scratch.refinenet2 = _make_fusion_block(features, device=device, dtype=dtype, operations=operations) + self.scratch.refinenet3 = _make_fusion_block(features, device=device, dtype=dtype, operations=operations) + self.scratch.refinenet4 = _make_fusion_block(features, has_residual=False, device=device, dtype=dtype, operations=operations) + # Auxiliary fusion chain (separate copies) + self.scratch.refinenet1_aux = _make_fusion_block(features, device=device, dtype=dtype, operations=operations) + self.scratch.refinenet2_aux = _make_fusion_block(features, device=device, dtype=dtype, operations=operations) + self.scratch.refinenet3_aux = _make_fusion_block(features, device=device, dtype=dtype, operations=operations) + self.scratch.refinenet4_aux = _make_fusion_block(features, has_residual=False, device=device, dtype=dtype, operations=operations) + + head_features_1 = features + head_features_2 = 32 + + # Main head neck + final projection + self.scratch.output_conv1 = operations.Conv2d( + head_features_1, head_features_1 // 2, kernel_size=3, stride=1, padding=1, + device=device, dtype=dtype, + ) + self.scratch.output_conv2 = nn.Sequential( + operations.Conv2d(head_features_1 // 2, head_features_2, kernel_size=3, stride=1, padding=1, device=device, dtype=dtype), + nn.ReLU(inplace=False), + operations.Conv2d(head_features_2, output_dim, kernel_size=1, stride=1, padding=0, device=device, dtype=dtype), + ) + + # Aux pre-head per level (multi-level pyramid) + self.scratch.output_conv1_aux = nn.ModuleList([ + self._make_aux_out1_block(head_features_1, device=device, dtype=dtype, operations=operations) + for _ in range(self.aux_levels) + ]) + + # Aux final projection per level (includes LayerNorm permute path). + ln_seq = [Permute((0, 2, 3, 1)), + operations.LayerNorm(head_features_2, device=device, dtype=dtype), + Permute((0, 3, 1, 2))] + self.scratch.output_conv2_aux = nn.ModuleList([ + nn.Sequential( + operations.Conv2d(head_features_1 // 2, head_features_2, kernel_size=3, stride=1, padding=1, device=device, dtype=dtype), + *ln_seq, + nn.ReLU(inplace=False), + operations.Conv2d(head_features_2, 7, kernel_size=1, stride=1, padding=0, device=device, dtype=dtype), + ) + for _ in range(self.aux_levels) + ]) + + @staticmethod + def _make_aux_out1_block(in_ch: int, *, device=None, dtype=None, operations=None) -> nn.Sequential: + # aux_out1_conv_num=5 in all Apache-2.0 variants. + return nn.Sequential( + operations.Conv2d(in_ch, in_ch // 2, 3, 1, 1, device=device, dtype=dtype), + operations.Conv2d(in_ch // 2, in_ch, 3, 1, 1, device=device, dtype=dtype), + operations.Conv2d(in_ch, in_ch // 2, 3, 1, 1, device=device, dtype=dtype), + operations.Conv2d(in_ch // 2, in_ch, 3, 1, 1, device=device, dtype=dtype), + operations.Conv2d(in_ch, in_ch // 2, 3, 1, 1, device=device, dtype=dtype), + ) + + def forward(self, feats: List[torch.Tensor], H: int, W: int, patch_start_idx: int = 0, **_kwargs) -> dict: + B, S, N, C = feats[0][0].shape + feats_flat = [feat[0].reshape(B * S, N, C) for feat in feats] + + ph, pw = H // self.patch_size, W // self.patch_size + resized = [] + for stage_idx, take_idx in enumerate(self.intermediate_layer_idx): + x = feats_flat[take_idx][:, patch_start_idx:] + x = self.norm(x) + x = x.permute(0, 2, 1).contiguous().reshape(B * S, C, ph, pw) + x = self.projects[stage_idx](x) + if self.pos_embed: + x = _add_pos_embed(x, W, H) + x = self.resize_layers[stage_idx](x) + resized.append(x) + + l1_rn = self.scratch.layer1_rn(resized[0]) + l2_rn = self.scratch.layer2_rn(resized[1]) + l3_rn = self.scratch.layer3_rn(resized[2]) + l4_rn = self.scratch.layer4_rn(resized[3]) + + # Main pyramid (output_conv1 is applied inside the upstream `_fuse`, + # before interpolation -- replicate that order here). + m = self.scratch.refinenet4(l4_rn, size=l3_rn.shape[2:]) + if self.enable_aux: + a4 = self.scratch.refinenet4_aux(l4_rn, size=l3_rn.shape[2:]) + aux_pyr = [a4] + m = self.scratch.refinenet3(m, l3_rn, size=l2_rn.shape[2:]) + if self.enable_aux: + aux_pyr.append(self.scratch.refinenet3_aux(aux_pyr[-1], l3_rn, size=l2_rn.shape[2:])) + m = self.scratch.refinenet2(m, l2_rn, size=l1_rn.shape[2:]) + if self.enable_aux: + aux_pyr.append(self.scratch.refinenet2_aux(aux_pyr[-1], l2_rn, size=l1_rn.shape[2:])) + m = self.scratch.refinenet1(m, l1_rn) + if self.enable_aux: + aux_pyr.append(self.scratch.refinenet1_aux(aux_pyr[-1], l1_rn)) + m = self.scratch.output_conv1(m) + + h_out = int(ph * self.patch_size / self.down_ratio) + w_out = int(pw * self.patch_size / self.down_ratio) + + m = _custom_interpolate(m, (h_out, w_out), mode="bilinear", align_corners=True) + if self.pos_embed: + m = _add_pos_embed(m, W, H) + main_logits = self.scratch.output_conv2(m) + fmap = main_logits.permute(0, 2, 3, 1) + depth_pred = _apply_activation(fmap[..., :-1], self.activation) + depth_conf = _apply_activation(fmap[..., -1], self.conf_activation) + + outs = { + self.head_main: depth_pred.squeeze(-1).view(B, S, *depth_pred.shape[1:-1]), + f"{self.head_main}_conf": depth_conf.view(B, S, *depth_conf.shape[1:]), + } + + if self.enable_aux: + # Auxiliary "ray" head (multi-level inside) -- only the last level + # is returned. Mirrors upstream ``DualDPT._fuse`` + ``_forward_impl``: + # each aux pyramid level goes through ``output_conv1_aux[i]`` + # (5-layer conv stack that ends at ``features // 2`` channels), + # then the last level optionally gets a pos-embed and finally + # ``output_conv2_aux[-1]``. + aux_processed = [ + self.scratch.output_conv1_aux[i](a) for i, a in enumerate(aux_pyr) + ] + last_aux = aux_processed[-1] + if self.pos_embed: + last_aux = _add_pos_embed(last_aux, W, H) + last_aux_logits = self.scratch.output_conv2_aux[-1](last_aux) + fmap_last = last_aux_logits.permute(0, 2, 3, 1) + # Channels: [ray(6), ray_conf(1)]; ray uses 'linear' activation. + aux_pred = fmap_last[..., :-1] + aux_conf = _apply_activation(fmap_last[..., -1], self.conf_activation) + outs[self.head_aux] = aux_pred.view(B, S, *aux_pred.shape[1:]) + outs[f"{self.head_aux}_conf"] = aux_conf.view(B, S, *aux_conf.shape[1:]) + + return outs diff --git a/comfy/ldm/depth_anything_3/model.py b/comfy/ldm/depth_anything_3/model.py new file mode 100644 index 000000000..f3c8a5ee3 --- /dev/null +++ b/comfy/ldm/depth_anything_3/model.py @@ -0,0 +1,236 @@ +from __future__ import annotations + +from typing import Dict, Optional, Sequence + +import torch +import torch.nn as nn + +from comfy.image_encoders.dino2 import Dinov2Model + +from .camera import CameraDec, CameraEnc +from .dpt import DPT, DualDPT +from .ray_pose import get_extrinsic_from_camray +from .transform import affine_inverse, pose_encoding_to_extri_intri + + +_HEAD_REGISTRY = { + "dpt": DPT, + "dualdpt": DualDPT, +} + + +# Backbone presets (mirror the upstream DINOv2 ViT variants). +_BACKBONE_PRESETS = { + "vits": dict(hidden_size=384, num_hidden_layers=12, num_attention_heads=6, use_swiglu_ffn=False), + "vitb": dict(hidden_size=768, num_hidden_layers=12, num_attention_heads=12, use_swiglu_ffn=False), + "vitl": dict(hidden_size=1024, num_hidden_layers=24, num_attention_heads=16, use_swiglu_ffn=False), + "vitg": dict(hidden_size=1536, num_hidden_layers=40, num_attention_heads=24, use_swiglu_ffn=True), +} + + +def _build_backbone_config( + backbone_name: str, + *, + alt_start: int, + qknorm_start: int, + rope_start: int, + cat_token: bool, +) -> dict: + if backbone_name not in _BACKBONE_PRESETS: + raise ValueError(f"Unknown DINOv2 backbone variant: {backbone_name!r}") + cfg = dict(_BACKBONE_PRESETS[backbone_name]) + cfg.update(dict( + layer_norm_eps=1e-6, + patch_size=14, + image_size=518, + # No mask_token in DA3 weights; omit param to avoid load warnings. + use_mask_token=False, + alt_start=alt_start, + qknorm_start=qknorm_start, + rope_start=rope_start, + cat_token=cat_token, + rope_freq=100.0, + )) + return cfg + + +class DepthAnything3Net(nn.Module): + + PATCH_SIZE = 14 + + def __init__( + self, + # --- Backbone --- + backbone_name: str = "vitl", + out_layers: Sequence[int] = (4, 11, 17, 23), + alt_start: int = -1, + qknorm_start: int = -1, + rope_start: int = -1, + cat_token: bool = False, + # --- Head --- + head_type: str = "dpt", # dpt or dualdpt + head_dim_in: int = 1024, + head_output_dim: int = 1, # 1 = depth only, 2 = depth+conf + head_features: int = 256, + head_out_channels: Sequence[int] = (256, 512, 1024, 1024), + head_use_sky_head: bool = True, # ignored by DualDPT + head_pos_embed: Optional[bool] = None, # default: True for DualDPT, False for DPT + # --- Camera (multi-view) --- + has_cam_enc: bool = False, + has_cam_dec: bool = False, + cam_dim_out: Optional[int] = None, # CameraEnc dim_out (defaults to embed_dim) + cam_dec_dim_in: Optional[int] = None, # CameraDec dim_in (defaults to 2*embed_dim with cat_token) + # ComfyUI plumbing + device=None, dtype=None, operations=None, + **_ignored, + ): + super().__init__() + head_cls = _HEAD_REGISTRY[head_type.lower()] + self.head_type = head_type.lower() + self.has_sky = (self.head_type == "dpt") and head_use_sky_head + self.has_conf = head_output_dim > 1 + self.out_layers = list(out_layers) + + backbone_cfg = _build_backbone_config( + backbone_name, + alt_start=alt_start, + qknorm_start=qknorm_start, + rope_start=rope_start, + cat_token=cat_token, + ) + self.backbone = Dinov2Model(backbone_cfg, dtype, device, operations) + + head_kwargs = dict( + dim_in=head_dim_in, + patch_size=self.PATCH_SIZE, + output_dim=head_output_dim, + features=head_features, + out_channels=tuple(head_out_channels), + device=device, dtype=dtype, operations=operations, + ) + if self.head_type == "dpt": + head_kwargs.update( + use_sky_head=head_use_sky_head, + pos_embed=(False if head_pos_embed is None else head_pos_embed), + ) + else: # dualdpt + head_kwargs.update( + pos_embed=(True if head_pos_embed is None else head_pos_embed), + ) + self.head = head_cls(**head_kwargs) + + # Built only if checkpoint has weights; cam_enc output dim == embed_dim. + embed_dim = backbone_cfg["hidden_size"] + if has_cam_enc: + self.cam_enc = CameraEnc( + dim_out=cam_dim_out if cam_dim_out is not None else embed_dim, + num_heads=max(1, embed_dim // 64), + device=device, dtype=dtype, operations=operations, + ) + else: + self.cam_enc = None + if has_cam_dec: + default_dim = embed_dim * (2 if cat_token else 1) + self.cam_dec = CameraDec( + dim_in=cam_dec_dim_in if cam_dec_dim_in is not None else default_dim, + device=device, dtype=dtype, operations=operations, + ) + else: + self.cam_dec = None + + self.dtype = dtype + + def forward( + self, + image: torch.Tensor, + extrinsics: Optional[torch.Tensor] = None, + intrinsics: Optional[torch.Tensor] = None, + *, + use_ray_pose: bool = False, + ref_view_strategy: str = "saddle_balanced", + export_feat_layers: Optional[Sequence[int]] = None, + **_unused, + ) -> Dict[str, torch.Tensor]: + """Run depth and optionally pose prediction.""" + if image.ndim == 4: + image = image.unsqueeze(1) # (B, 1, 3, H, W) + assert image.ndim == 5 and image.shape[2] == 3, \ + f"image must be (B,3,H,W) or (B,S,3,H,W); got {tuple(image.shape)}" + + B, S, _, H, W = image.shape + assert H % self.PATCH_SIZE == 0 and W % self.PATCH_SIZE == 0, \ + f"image H,W must be multiples of {self.PATCH_SIZE}; got {(H, W)}" + + # Camera-token preparation (multi-view path). + cam_token = None + if extrinsics is not None and intrinsics is not None and self.cam_enc is not None: + cam_token = self.cam_enc(extrinsics, intrinsics, (H, W)) + + # Toggle aux ray output on/off depending on what the caller asked for. + if isinstance(self.head, DualDPT): + self.head.enable_aux = bool(use_ray_pose) + + feats, aux_feats = self.backbone.get_intermediate_layers_da3( + image, self.out_layers, cam_token=cam_token, + ref_view_strategy=ref_view_strategy, + export_feat_layers=export_feat_layers, + ) + head_out = self.head(feats, H=H, W=W, patch_start_idx=0) + + # Pose prediction. + out: Dict[str, torch.Tensor] = {} + if use_ray_pose and "ray" in head_out and "ray_conf" in head_out: + ray = head_out["ray"] + ray_conf = head_out["ray_conf"] + extr_c2w, focal, pp = get_extrinsic_from_camray( + ray, ray_conf, ray.shape[-3], ray.shape[-2], + ) + # Match the upstream output: w2c, drop the homogeneous row. + extr_w2c = affine_inverse(extr_c2w)[:, :, :3, :] + # Build pixel-space intrinsics from the normalised focal/pp output. + intr = torch.eye(3, device=ray.device, dtype=ray.dtype) + intr = intr[None, None].expand(extr_c2w.shape[0], extr_c2w.shape[1], 3, 3).clone() + intr[:, :, 0, 0] = focal[:, :, 0] / 2 * W + intr[:, :, 1, 1] = focal[:, :, 1] / 2 * H + intr[:, :, 0, 2] = pp[:, :, 0] * W * 0.5 + intr[:, :, 1, 2] = pp[:, :, 1] * H * 0.5 + out["extrinsics"] = extr_w2c + out["intrinsics"] = intr + elif self.cam_dec is not None and S > 1: + # Decode the cam-token of the final out_layer into a pose encoding. + cam_feat = feats[-1][1] # (B, S, dim_in_to_cam_dec) + pose_enc = self.cam_dec(cam_feat) + c2w_3x4, intr = pose_encoding_to_extri_intri(pose_enc, (H, W)) + # Match the upstream output convention: w2c (world->camera), 3x4. + c2w_4x4 = torch.cat([ + c2w_3x4, + torch.tensor([0, 0, 0, 1], device=c2w_3x4.device, dtype=c2w_3x4.dtype) + .view(1, 1, 1, 4).expand(B, S, 1, 4), + ], dim=-2) + out["extrinsics"] = affine_inverse(c2w_4x4)[:, :, :3, :] + out["intrinsics"] = intr + + # Flatten the views axis for per-pixel outputs (depth/conf/sky) so the + # per-image consumer keeps its (B*S, H, W) interface. + for k, v in head_out.items(): + if k in ("ray", "ray_conf"): + # Keep multi-view shape for downstream pose work. + out[k] = v + elif v.ndim >= 3 and v.shape[0] == B and v.shape[1] == S: + out[k] = v.reshape(B * S, *v.shape[2:]) + else: + out[k] = v + + if export_feat_layers: + out["aux_features"] = self._reshape_aux_features(aux_feats, H, W) + return out + + def _reshape_aux_features(self, aux_feats, H: int, W: int): + """Reshape (B, S, N, C) aux features into (B, S, h_p, w_p, C).""" + ph, pw = H // self.PATCH_SIZE, W // self.PATCH_SIZE + out = [] + for f in aux_feats: + B, S, N, C = f.shape + assert N == ph * pw, f"aux feature seq mismatch: {N} != {ph}*{pw}" + out.append(f.reshape(B, S, ph, pw, C)) + return out diff --git a/comfy/ldm/depth_anything_3/preprocess.py b/comfy/ldm/depth_anything_3/preprocess.py new file mode 100644 index 000000000..2238bd0d6 --- /dev/null +++ b/comfy/ldm/depth_anything_3/preprocess.py @@ -0,0 +1,128 @@ +"""Input/output preprocessing helpers for Depth Anything 3.""" + +from __future__ import annotations + +from typing import Tuple + +import torch + +import comfy.utils + +PATCH_SIZE = 14 + +# ImageNet normalization constants used during DA3 training. +_IMAGENET_MEAN = torch.tensor([0.485, 0.456, 0.406]) +_IMAGENET_STD = torch.tensor([0.229, 0.224, 0.225]) + + +def _round_to_patch(x: int, patch: int = PATCH_SIZE) -> int: + down = (x // patch) * patch + up = down + patch + return up if abs(up - x) <= abs(x - down) else down + + +def compute_target_size(orig_h: int, orig_w: int, process_res: int, method: str = "upper_bound_resize") -> Tuple[int, int]: + """Compute (target_h, target_w) for a single image. + upper_bound_resize: scale longest side to process_res, then round each dim to nearest multiple of 14 (default upstream method). + lower_bound_resize: scale shortest side to process_res, then round.""" + + if method == "upper_bound_resize": + longest = max(orig_h, orig_w) + scale = process_res / float(longest) + elif method == "lower_bound_resize": + shortest = min(orig_h, orig_w) + scale = process_res / float(shortest) + else: + raise ValueError(f"Unsupported process_res_method: {method}") + + new_w = max(1, _round_to_patch(int(round(orig_w * scale)))) + new_h = max(1, _round_to_patch(int(round(orig_h * scale)))) + return new_h, new_w + + +def preprocess_image(image: torch.Tensor, process_res: int = 504, method: str = "upper_bound_resize") -> torch.Tensor: + assert image.ndim == 4 and image.shape[-1] == 3, f"expected (B,H,W,3) IMAGE; got {tuple(image.shape)}" + B, H, W, _ = image.shape + target_h, target_w = compute_target_size(H, W, process_res, method) + + # (B, H, W, 3) -> (B, 3, H, W) + x = image.movedim(-1, 1).contiguous() + if (target_h, target_w) != (H, W): + # Upstream uses cv2 INTER_CUBIC (upscale) / INTER_AREA (downscale). + # Lanczos in ``common_upscale`` is anti-aliased and produces the + # closest pixel-wise match in a sweep across {bilinear, bicubic, + # area, lanczos, bislerp}. Used in both directions for simplicity. + x = comfy.utils.common_upscale(x.float(), target_w, target_h, "lanczos", "disabled",) + x = x.clamp(0.0, 1.0) + + mean = _IMAGENET_MEAN.to(device=x.device, dtype=x.dtype).view(1, 3, 1, 1) + std = _IMAGENET_STD.to(device=x.device, dtype=x.dtype).view(1, 3, 1, 1) + x = (x - mean) / std + return x + + +# ----------------------------------------------------------------------------- +# Output post-processing (sky-aware clipping for Mono/Metric variants) +# ----------------------------------------------------------------------------- + + +def compute_non_sky_mask(sky_prediction: torch.Tensor, threshold: float = 0.3) -> torch.Tensor: + """Boolean mask: True for non-sky pixels (sky probability < threshold).""" + return sky_prediction < threshold + + +def apply_sky_aware_clip(depth: torch.Tensor, sky: torch.Tensor, threshold: float = 0.3, quantile: float = 0.99) -> torch.Tensor: + """Clips sky regions to the 99th percentile of non-sky depth. Returns a new depth tensor.""" + non_sky = compute_non_sky_mask(sky, threshold=threshold) + if non_sky.sum() <= 10 or (~non_sky).sum() <= 10: + return depth.clone() + + non_sky_depth = depth[non_sky] + if non_sky_depth.numel() > 100_000: + idx = torch.randint(0, non_sky_depth.numel(), (100_000,), device=non_sky_depth.device) + sampled = non_sky_depth[idx] + else: + sampled = non_sky_depth + + max_depth = torch.quantile(sampled, quantile) + out = depth.clone() + out[~non_sky] = max_depth + return out + + +def normalize_depth_v2_style(depth: torch.Tensor, sky: torch.Tensor | None = None, low_quantile: float = 0.01, high_quantile: float = 0.99) -> torch.Tensor: + """V2-style normalization computes percentile bounds over non-sky pixels (when available), then maps depth into [0, 1] with near = white (1.0).""" + if sky is not None: + mask = compute_non_sky_mask(sky) + if mask.any(): + valid = depth[mask] + else: + valid = depth.flatten() + else: + valid = depth.flatten() + + if valid.numel() > 100_000: + idx = torch.randint(0, valid.numel(), (100_000,), device=valid.device) + sample = valid[idx] + else: + sample = valid + + lo = torch.quantile(sample, low_quantile) + hi = torch.quantile(sample, high_quantile) + rng = (hi - lo).clamp(min=1e-6) + norm = ((depth - lo) / rng).clamp(0.0, 1.0) + # Nearer pixels are brighter (1.0) + norm = 1.0 - norm + if sky is not None: + # Sky pixels become black (far / unknown) + sky_mask = ~compute_non_sky_mask(sky) + norm = torch.where(sky_mask, torch.zeros_like(norm), norm) + return norm + + +def normalize_depth_min_max(depth: torch.Tensor) -> torch.Tensor: + """Simple per-frame min/max normalization with near=1.0 convention.""" + lo = depth.amin(dim=(-2, -1), keepdim=True) + hi = depth.amax(dim=(-2, -1), keepdim=True) + rng = (hi - lo).clamp(min=1e-6) + return 1.0 - ((depth - lo) / rng).clamp(0.0, 1.0) diff --git a/comfy/ldm/depth_anything_3/ray_pose.py b/comfy/ldm/depth_anything_3/ray_pose.py new file mode 100644 index 000000000..90890f1da --- /dev/null +++ b/comfy/ldm/depth_anything_3/ray_pose.py @@ -0,0 +1,272 @@ +"""Ray-to-pose conversion for the multi-view path of Depth Anything 3.""" + +from __future__ import annotations + +from typing import Optional, Tuple + +import torch + + +# qr/svd use fp32: CUDA often has no fp16/bf16 kernels for these ops. + + +def _ql_decomposition(A: torch.Tensor) -> Tuple[torch.Tensor, torch.Tensor]: + """Decompose A = Q @ L with Q orthogonal and L lower-triangular. + Implemented in terms of QR by reversing the columns/rows; the standard + trick from the upstream reference. Inputs A are (3, 3).""" + P = torch.tensor([[0, 0, 1], [0, 1, 0], [1, 0, 0]], device=A.device, dtype=A.dtype) + A_tilde = A @ P + # CUDA QR is not implemented for fp16/bf16; upcast just for this call. + Q_tilde, R_tilde = torch.linalg.qr(A_tilde.float()) + Q_tilde = Q_tilde.to(A.dtype) + R_tilde = R_tilde.to(A.dtype) + Q = Q_tilde @ P + L = P @ R_tilde @ P + d = torch.diag(L) + sign = torch.sign(d) + Q = Q * sign[None, :] # scale columns of Q + L = L * sign[:, None] # scale rows of L + return Q, L + + +def _homogenize_points(points: torch.Tensor) -> torch.Tensor: + return torch.cat([points, torch.ones_like(points[..., :1])], dim=-1) + + +# ----------------------------------------------------------------------------- +# Weighted-LSQ + RANSAC homography (batched) +# ----------------------------------------------------------------------------- + + +def _find_homography_weighted_lsq(src_pts: torch.Tensor, dst_pts: torch.Tensor, confident_weight: torch.Tensor,) -> torch.Tensor: + """Solve a single H with weighted least-squares (DLT).""" + N = src_pts.shape[0] + if N < 4: + raise ValueError("At least 4 points are required to compute a homography.") + w = confident_weight.sqrt().unsqueeze(1) # (N, 1) + x = src_pts[:, 0:1] + y = src_pts[:, 1:2] + u = dst_pts[:, 0:1] + v = dst_pts[:, 1:2] + zeros = torch.zeros_like(x) + A1 = torch.cat([-x * w, -y * w, -w, zeros, zeros, zeros, x * u * w, y * u * w, u * w], dim=1) + A2 = torch.cat([zeros, zeros, zeros, -x * w, -y * w, -w, x * v * w, y * v * w, v * w], dim=1) + A = torch.cat([A1, A2], dim=0) # (2N, 9) + # CUDA SVD is not implemented for fp16/bf16; upcast just for this call. + _, _, Vh = torch.linalg.svd(A.float()) + Vh = Vh.to(A.dtype) + H = Vh[-1].reshape(3, 3) + return H / H[-1, -1] + + +def _find_homography_weighted_lsq_batched(src_pts_batch: torch.Tensor, dst_pts_batch: torch.Tensor, confident_weight_batch: torch.Tensor) -> torch.Tensor: + """Batched DLT solver. Inputs (B, K, 2) / (B, K); output (B, 3, 3).""" + B, K, _ = src_pts_batch.shape + w = confident_weight_batch.sqrt().unsqueeze(2) + x = src_pts_batch[:, :, 0:1] + y = src_pts_batch[:, :, 1:2] + u = dst_pts_batch[:, :, 0:1] + v = dst_pts_batch[:, :, 1:2] + zeros = torch.zeros_like(x) + A1 = torch.cat([-x * w, -y * w, -w, zeros, zeros, zeros, x * u * w, y * u * w, u * w], dim=2) + A2 = torch.cat([zeros, zeros, zeros, -x * w, -y * w, -w, x * v * w, y * v * w, v * w], dim=2) + A = torch.cat([A1, A2], dim=1) # (B, 2K, 9) + # CUDA SVD is not implemented for fp16/bf16; upcast just for this call. + _, _, Vh = torch.linalg.svd(A.float()) + Vh = Vh.to(A.dtype) + H = Vh[:, -1].reshape(B, 3, 3) + return H / H[:, 2:3, 2:3] + + +def _ransac_find_homography_weighted_batched( + src_pts: torch.Tensor, # (B, N, 2) + dst_pts: torch.Tensor, # (B, N, 2) + confident_weight: torch.Tensor, # (B, N) + n_sample: int, + n_iter: int = 100, + reproj_threshold: float = 3.0, + num_sample_for_ransac: int = 8, + max_inlier_num: int = 10000, + rand_sample_iters_idx: Optional[torch.Tensor] = None, +) -> torch.Tensor: + """Batched weighted-RANSAC homography estimator. Returns (B, 3, 3) homography matrices.""" + B, N, _ = src_pts.shape + assert N >= 4 + device = src_pts.device + + sorted_idx = torch.argsort(confident_weight, descending=True, dim=1) + candidate_idx = sorted_idx[:, :n_sample] # (B, n_sample) + + if rand_sample_iters_idx is None: + rand_sample_iters_idx = torch.stack( + [torch.randperm(n_sample, device=device)[:num_sample_for_ransac] + for _ in range(n_iter)], + dim=0, + ) + + rand_idx = candidate_idx[:, rand_sample_iters_idx] # (B, n_iter, k) + b_idx = ( + torch.arange(B, device=device) + .view(B, 1, 1) + .expand(B, n_iter, num_sample_for_ransac) + ) + src_b = src_pts[b_idx, rand_idx] + dst_b = dst_pts[b_idx, rand_idx] + w_b = confident_weight[b_idx, rand_idx] + + cB, cN = src_b.shape[:2] + H_batch = _find_homography_weighted_lsq_batched( + src_b.flatten(0, 1), dst_b.flatten(0, 1), w_b.flatten(0, 1), + ).unflatten(0, (cB, cN)) # (B, n_iter, 3, 3) + + src_homo = torch.cat([src_pts, torch.ones(B, N, 1, device=device, dtype=src_pts.dtype)], dim=2) + proj = torch.bmm( + src_homo.unsqueeze(1).expand(B, n_iter, N, 3).reshape(-1, N, 3), + H_batch.reshape(-1, 3, 3).transpose(1, 2), + ) # (B*n_iter, N, 3) + proj_xy = (proj[:, :, :2] / proj[:, :, 2:3]).reshape(B, n_iter, N, 2) + err = ((proj_xy - dst_pts.unsqueeze(1)) ** 2).sum(-1).sqrt() # (B, n_iter, N) + inlier_mask = err < reproj_threshold + score = (inlier_mask * confident_weight.unsqueeze(1)).sum(dim=2) + best_idx = torch.argmax(score, dim=1) + best_inlier_mask = inlier_mask[torch.arange(B, device=device), best_idx] + + # Refit with the inlier set (per-batch, since the inlier counts vary). + H_inlier_list = [] + for b in range(B): + mask = best_inlier_mask[b] + in_src = src_pts[b][mask] + in_dst = dst_pts[b][mask] + in_w = confident_weight[b][mask] + if in_src.shape[0] < 4: + # Fall back to identity when RANSAC fails to find enough inliers. + H_inlier_list.append(torch.eye(3, device=device, dtype=src_pts.dtype)) + continue + sorted_w = torch.argsort(in_w, descending=True) + if len(sorted_w) > max_inlier_num: + keep = max(int(len(sorted_w) * 0.95), max_inlier_num) + sorted_w = sorted_w[:keep][torch.randperm(keep, device=device)[:max_inlier_num]] + H_inlier_list.append( + _find_homography_weighted_lsq(in_src[sorted_w], in_dst[sorted_w], in_w[sorted_w]) + ) + return torch.stack(H_inlier_list, dim=0) + + +# ----------------------------------------------------------------------------- +# Camera-ray utilities +# ----------------------------------------------------------------------------- + + +def _unproject_identity(num_y: int, num_x: int, B: int, S: int, device, dtype) -> torch.Tensor: + """Camera-space unit rays for an identity intrinsic on a 2x2 image plane.""" + dx = 1.0 / num_x + dy = 1.0 / num_y + # Centered camera-space coords directly (skip the K^-1 step since it's + # just a translation by -1 on x and y when K is identity-with-center=1). + y = torch.linspace(-(1 - dy), (1 - dy), num_y, device=device, dtype=dtype) + x = torch.linspace(-(1 - dx), (1 - dx), num_x, device=device, dtype=dtype) + yy, xx = torch.meshgrid(y, x, indexing="ij") + grid = torch.stack((xx, yy), dim=-1) # (h, w, 2) + grid = grid.unsqueeze(0).unsqueeze(0).expand(B, S, num_y, num_x, 2) + return torch.cat([grid, torch.ones_like(grid[..., :1])], dim=-1) + + +def _camray_to_caminfo( + camray: torch.Tensor, # (B, S, h, w, 6) + confidence: Optional[torch.Tensor] = None, # (B, S, h, w) + reproj_threshold: float = 0.2, +) -> Tuple[torch.Tensor, torch.Tensor, torch.Tensor, torch.Tensor]: + """Convert per-pixel camera rays to per-view (R, T, focal, principal).""" + if confidence is None: + confidence = torch.ones_like(camray[..., 0]) + B, S, h, w, _ = camray.shape + device = camray.device + dtype = camray.dtype + + rays_target = camray[..., :3] # (B, S, h, w, 3) + rays_origin = _unproject_identity(h, w, B, S, device, dtype) + + # Flatten (B*S, h*w, *) for the RANSAC routine. + rays_target = rays_target.flatten(0, 1).flatten(1, 2) + rays_origin = rays_origin.flatten(0, 1).flatten(1, 2) + weights = confidence.flatten(0, 1).flatten(1, 2).clone() + + # Project to 2D in homogeneous form (the upstream calls this "perspective division"). + z_thresh = 1e-4 + mask = (rays_target[:, :, 2].abs() > z_thresh) & (rays_origin[:, :, 2].abs() > z_thresh) + weights = torch.where(mask, weights, torch.zeros_like(weights)) + src = rays_origin.clone() + dst = rays_target.clone() + src[..., 0] = torch.where(mask, src[..., 0] / src[..., 2], src[..., 0]) + src[..., 1] = torch.where(mask, src[..., 1] / src[..., 2], src[..., 1]) + dst[..., 0] = torch.where(mask, dst[..., 0] / dst[..., 2], dst[..., 0]) + dst[..., 1] = torch.where(mask, dst[..., 1] / dst[..., 2], dst[..., 1]) + src = src[..., :2] + dst = dst[..., :2] + + N = src.shape[1] + n_iter = 100 + sample_ratio = 0.3 + num_sample_for_ransac = 8 + n_sample = max(num_sample_for_ransac, int(N * sample_ratio)) + rand_idx = torch.stack( + [torch.randperm(n_sample, device=device)[:num_sample_for_ransac] for _ in range(n_iter)], + dim=0, + ) + + # Chunk along the view axis to keep peak memory predictable. + chunk = 2 + A_list = [] + for i in range(0, src.shape[0], chunk): + A = _ransac_find_homography_weighted_batched( + src[i:i + chunk], dst[i:i + chunk], weights[i:i + chunk], + n_sample=n_sample, n_iter=n_iter, + num_sample_for_ransac=num_sample_for_ransac, + reproj_threshold=reproj_threshold, + rand_sample_iters_idx=rand_idx, + max_inlier_num=8000, + ) + # Flip sign on dets that come out < 0 (so that the QL produces a + # right-handed rotation). ``det`` lacks fp16/bf16 CUDA kernels, so + # do the comparison in fp32. + flip = torch.linalg.det(A.float()) < 0 + A = torch.where(flip[:, None, None], -A, A) + A_list.append(A) + A = torch.cat(A_list, dim=0) # (B*S, 3, 3) + + R_list, f_list, pp_list = [], [], [] + for i in range(A.shape[0]): + R, L = _ql_decomposition(A[i]) + L = L / L[2][2] + f_list.append(torch.stack((L[0][0], L[1][1]))) + pp_list.append(torch.stack((L[2][0], L[2][1]))) + R_list.append(R) + R = torch.stack(R_list).reshape(B, S, 3, 3) + focal = torch.stack(f_list).reshape(B, S, 2) + pp = torch.stack(pp_list).reshape(B, S, 2) + + # Translation: confidence-weighted average of camray direction(s). + cf = confidence.flatten(0, 1).flatten(1, 2) + T = (camray.flatten(0, 1).flatten(1, 2)[..., 3:] * cf.unsqueeze(-1)).sum(dim=1) + T = T / cf.sum(dim=-1, keepdim=True) + T = T.reshape(B, S, 3) + + # Match upstream output convention: focal -> 1/focal, pp + 1. + return R, T, 1.0 / focal, pp + 1.0 + + +def get_extrinsic_from_camray( + camray: torch.Tensor, # (B, S, h, w, 6) + conf: torch.Tensor, # (B, S, h, w, 1) or (B, S, h, w) + patch_size_y: int, + patch_size_x: int, +) -> Tuple[torch.Tensor, torch.Tensor, torch.Tensor]: + """Wrap a 4x4 extrinsic + per-view focal + principal-point output.""" + if conf.ndim == 5 and conf.shape[-1] == 1: + conf = conf.squeeze(-1) + R, T, focal, pp = _camray_to_caminfo(camray, confidence=conf) + extr = torch.cat([R, T.unsqueeze(-1)], dim=-1) # (B, S, 3, 4) + homo_row = torch.tensor([0, 0, 0, 1], dtype=R.dtype, device=R.device) + homo_row = homo_row.view(1, 1, 1, 4).expand(R.shape[0], R.shape[1], 1, 4) + extr = torch.cat([extr, homo_row], dim=-2) # (B, S, 4, 4) + return extr, focal, pp diff --git a/comfy/ldm/depth_anything_3/reference_view_selector.py b/comfy/ldm/depth_anything_3/reference_view_selector.py new file mode 100644 index 000000000..90f00be92 --- /dev/null +++ b/comfy/ldm/depth_anything_3/reference_view_selector.py @@ -0,0 +1,87 @@ +"""Reference-view selection for the multi-view path of Depth Anything 3.""" + +from __future__ import annotations + +from typing import Literal + +import torch + + +RefViewStrategy = Literal["first", "middle", "saddle_balanced", "saddle_sim_range"] + + +# Per the upstream constants module: ``THRESH_FOR_REF_SELECTION = 3``. +# Reference selection only runs when there are at least this many views. +THRESH_FOR_REF_SELECTION: int = 3 + + +def select_reference_view(x: torch.Tensor, strategy: RefViewStrategy = "saddle_balanced") -> torch.Tensor: + """Pick a reference view index per batch element.""" + B, S, _, _ = x.shape + if S <= 1: + return torch.zeros(B, dtype=torch.long, device=x.device) + if strategy == "first": + return torch.zeros(B, dtype=torch.long, device=x.device) + if strategy == "middle": + return torch.full((B,), S // 2, dtype=torch.long, device=x.device) + + # Feature-based strategies: normalised cls/cam token per view. + img_class_feat = x[:, :, 0] / x[:, :, 0].norm(dim=-1, keepdim=True) # (B,S,C) + + if strategy == "saddle_balanced": + sim = torch.matmul(img_class_feat, img_class_feat.transpose(1, 2)) # (B,S,S) + sim_no_diag = sim - torch.eye(S, device=sim.device).unsqueeze(0) + sim_score = sim_no_diag.sum(dim=-1) / (S - 1) # (B,S) + feat_norm = x[:, :, 0].norm(dim=-1) # (B,S) + feat_var = img_class_feat.var(dim=-1) # (B,S) + + def _normalize(metric): + mn = metric.min(dim=1, keepdim=True).values + mx = metric.max(dim=1, keepdim=True).values + return (metric - mn) / (mx - mn + 1e-8) + + sim_n, norm_n, var_n = _normalize(sim_score), _normalize(feat_norm), _normalize(feat_var) + balance = (sim_n - 0.5).abs() + (norm_n - 0.5).abs() + (var_n - 0.5).abs() + return balance.argmin(dim=1) + + if strategy == "saddle_sim_range": + sim = torch.matmul(img_class_feat, img_class_feat.transpose(1, 2)) + sim_no_diag = sim - torch.eye(S, device=sim.device).unsqueeze(0) + sim_max = sim_no_diag.max(dim=-1).values + sim_min = sim_no_diag.min(dim=-1).values + return (sim_max - sim_min).argmax(dim=1) + + raise ValueError( + f"Unknown reference view selection strategy: {strategy!r}. " + f"Must be one of: 'first', 'middle', 'saddle_balanced', 'saddle_sim_range'" + ) + + +def reorder_by_reference(x: torch.Tensor, b_idx: torch.Tensor) -> torch.Tensor: + """Reorder x so the reference view is at position 0 in axis S.""" + B, S = x.shape[0], x.shape[1] + if S <= 1: + return x + positions = torch.arange(S, device=x.device).unsqueeze(0).expand(B, -1) + b_idx_exp = b_idx.unsqueeze(1) + reorder = torch.where( + (positions > 0) & (positions <= b_idx_exp), + positions - 1, + positions, + ) + reorder[:, 0] = b_idx + batch = torch.arange(B, device=x.device).unsqueeze(1) + return x[batch, reorder] + + +def restore_original_order(x: torch.Tensor, b_idx: torch.Tensor) -> torch.Tensor: + """Inverse of reorder_by_reference.""" + B, S = x.shape[0], x.shape[1] + if S <= 1: + return x + target_positions = torch.arange(S, device=x.device).unsqueeze(0).expand(B, -1) + b_idx_exp = b_idx.unsqueeze(1) + restore = torch.where(target_positions < b_idx_exp, target_positions + 1, target_positions) + restore = torch.scatter(restore, dim=1, index=b_idx_exp, src=torch.zeros_like(b_idx_exp)) + batch = torch.arange(B, device=x.device).unsqueeze(1) + return x[batch, restore] diff --git a/comfy/ldm/depth_anything_3/transform.py b/comfy/ldm/depth_anything_3/transform.py new file mode 100644 index 000000000..b735d7bec --- /dev/null +++ b/comfy/ldm/depth_anything_3/transform.py @@ -0,0 +1,160 @@ +"""Geometry / camera transform helpers for Depth Anything 3.""" + +from __future__ import annotations + +from typing import Tuple + +import torch +import torch.nn.functional as F + + +# ----------------------------------------------------------------------------- +# Affine 4x4 helpers +# ----------------------------------------------------------------------------- + + +def as_homogeneous(ext: torch.Tensor) -> torch.Tensor: + """Promote (...,3,4) extrinsics to (...,4,4) homogeneous form. No-op when the input is already ``(...,4,4)``.""" + if ext.shape[-2:] == (4, 4): + return ext + if ext.shape[-2:] == (3, 4): + ones = torch.zeros_like(ext[..., :1, :4]) + ones[..., 0, 3] = 1.0 + return torch.cat([ext, ones], dim=-2) + raise ValueError(f"Invalid affine shape: {ext.shape}") + + +def affine_inverse(A: torch.Tensor) -> torch.Tensor: + """Inverse of an affine matrix ``[R|T; 0 0 0 1]``.""" + R = A[..., :3, :3] + T = A[..., :3, 3:] + P = A[..., 3:, :] + return torch.cat([torch.cat([R.mT, -R.mT @ T], dim=-1), P], dim=-2) + + +# ----------------------------------------------------------------------------- +# Quaternion <-> rotation matrix (xyzw / scalar-last) +# ----------------------------------------------------------------------------- + + +def _sqrt_positive_part(x: torch.Tensor) -> torch.Tensor: + """sqrt(max(0, x)) with a zero subgradient where x == 0.""" + ret = torch.zeros_like(x) + positive_mask = x > 0 + if torch.is_grad_enabled(): + ret[positive_mask] = torch.sqrt(x[positive_mask]) + else: + ret = torch.where(positive_mask, torch.sqrt(x), ret) + return ret + + +def standardize_quaternion(quaternions: torch.Tensor) -> torch.Tensor: + """Force the real part of a unit quaternion (xyzw) to be non-negative.""" + return torch.where(quaternions[..., 3:4] < 0, -quaternions, quaternions) + + +def quat_to_mat(quaternions: torch.Tensor) -> torch.Tensor: + """Convert quaternions (xyzw) to (...,3,3) rotation matrices.""" + i, j, k, r = torch.unbind(quaternions, -1) + two_s = 2.0 / (quaternions * quaternions).sum(-1) + o = torch.stack( + ( + 1 - two_s * (j * j + k * k), + two_s * (i * j - k * r), + two_s * (i * k + j * r), + two_s * (i * j + k * r), + 1 - two_s * (i * i + k * k), + two_s * (j * k - i * r), + two_s * (i * k - j * r), + two_s * (j * k + i * r), + 1 - two_s * (i * i + j * j), + ), + -1, + ) + return o.reshape(quaternions.shape[:-1] + (3, 3)) + + +def mat_to_quat(matrix: torch.Tensor) -> torch.Tensor: + """Convert (...,3,3) rotation matrices to quaternions (xyzw).""" + if matrix.size(-1) != 3 or matrix.size(-2) != 3: + raise ValueError(f"Invalid rotation matrix shape {matrix.shape}.") + + batch_dim = matrix.shape[:-2] + m00, m01, m02, m10, m11, m12, m20, m21, m22 = torch.unbind( + matrix.reshape(batch_dim + (9,)), dim=-1 + ) + + q_abs = _sqrt_positive_part( + torch.stack( + [ + 1.0 + m00 + m11 + m22, + 1.0 + m00 - m11 - m22, + 1.0 - m00 + m11 - m22, + 1.0 - m00 - m11 + m22, + ], + dim=-1, + ) + ) + + quat_by_rijk = torch.stack( + [ + torch.stack([q_abs[..., 0] ** 2, m21 - m12, m02 - m20, m10 - m01], dim=-1), + torch.stack([m21 - m12, q_abs[..., 1] ** 2, m10 + m01, m02 + m20], dim=-1), + torch.stack([m02 - m20, m10 + m01, q_abs[..., 2] ** 2, m12 + m21], dim=-1), + torch.stack([m10 - m01, m20 + m02, m21 + m12, q_abs[..., 3] ** 2], dim=-1), + ], + dim=-2, + ) + + flr = torch.tensor(0.1).to(dtype=q_abs.dtype, device=q_abs.device) + quat_candidates = quat_by_rijk / (2.0 * q_abs[..., None].max(flr)) + + out = quat_candidates[F.one_hot(q_abs.argmax(dim=-1), num_classes=4) > 0.5, :].reshape( + batch_dim + (4,) + ) + # Reorder rijk -> xyzw (i.e. ijkr). + out = out[..., [1, 2, 3, 0]] + return standardize_quaternion(out) + + +# ----------------------------------------------------------------------------- +# Pose-encoding <-> extrinsics + intrinsics +# ----------------------------------------------------------------------------- + + +def extri_intri_to_pose_encoding(extrinsics: torch.Tensor, intrinsics: torch.Tensor, image_size_hw: Tuple[int, int]) -> torch.Tensor: + """Pack (extr, intr, image_size) into the 9-D pose-encoding vector. + extrinsics: camera-to-world (c2w) (B,S,4,4) matrices, + intrinsics: pixel-space (B,S,3,3) matrices, + image_size_hw: is a (H, W) pair. + """ + R = extrinsics[..., :3, :3] + T = extrinsics[..., :3, 3] + quat = mat_to_quat(R) + H, W = image_size_hw + fov_h = 2 * torch.atan((H / 2) / intrinsics[..., 1, 1]) + fov_w = 2 * torch.atan((W / 2) / intrinsics[..., 0, 0]) + return torch.cat([T, quat, fov_h[..., None], fov_w[..., None]], dim=-1).float() + + +def pose_encoding_to_extri_intri(pose_encoding: torch.Tensor, image_size_hw: Tuple[int, int]) -> Tuple[torch.Tensor, torch.Tensor]: + """Inverse of extri_intri_to_pose_encoding.""" + T = pose_encoding[..., :3] + quat = pose_encoding[..., 3:7] + fov_h = pose_encoding[..., 7] + fov_w = pose_encoding[..., 8] + # Normalize to unit quaternion. CameraDec outputs raw values; a near-zero + # quaternion causes two_s = 2/norm² → inf in quat_to_mat → NaN extrinsics. + quat = quat / quat.norm(dim=-1, keepdim=True).clamp(min=1e-6) + R = quat_to_mat(quat) + extrinsics = torch.cat([R, T[..., None]], dim=-1) + H, W = image_size_hw + fy = (H / 2.0) / torch.clamp(torch.tan(fov_h / 2.0), 1e-6) + fx = (W / 2.0) / torch.clamp(torch.tan(fov_w / 2.0), 1e-6) + intrinsics = torch.zeros(pose_encoding.shape[:2] + (3, 3), device=pose_encoding.device, dtype=pose_encoding.dtype) + intrinsics[..., 0, 0] = fx + intrinsics[..., 1, 1] = fy + intrinsics[..., 0, 2] = W / 2 + intrinsics[..., 1, 2] = H / 2 + intrinsics[..., 2, 2] = 1.0 + return extrinsics, intrinsics diff --git a/comfy/model_base.py b/comfy/model_base.py index 2a46d1fc1..2289e0812 100644 --- a/comfy/model_base.py +++ b/comfy/model_base.py @@ -65,6 +65,7 @@ import comfy.ldm.ernie.model import comfy.ldm.sam3.detector import comfy.ldm.hidream_o1.model from comfy.ldm.hidream_o1.conditioning import build_extra_conds +import comfy.ldm.depth_anything_3.model import comfy.model_management import comfy.patcher_extension @@ -2319,6 +2320,12 @@ class RT_DETR_v4(BaseModel): def __init__(self, model_config, model_type=ModelType.FLOW, device=None): super().__init__(model_config, model_type, device=device, unet_model=comfy.ldm.rt_detr.rtdetr_v4.RTv4) + +class DepthAnything3(BaseModel): + def __init__(self, model_config, model_type=ModelType.FLOW, device=None): + super().__init__(model_config, model_type, device=device, + unet_model=comfy.ldm.depth_anything_3.model.DepthAnything3Net) + class ErnieImage(BaseModel): def __init__(self, model_config, model_type=ModelType.FLOW, device=None): super().__init__(model_config, model_type, device=device, unet_model=comfy.ldm.ernie.model.ErnieImageModel) diff --git a/comfy/model_detection.py b/comfy/model_detection.py index 290938bd6..7d0cab308 100644 --- a/comfy/model_detection.py +++ b/comfy/model_detection.py @@ -862,6 +862,95 @@ def detect_unet_config(state_dict, key_prefix, metadata=None): dit_config["enc_h"] = state_dict['{}encoder.pan_blocks.1.cv4.conv.weight'.format(key_prefix)].shape[0] return dit_config + # Depth Anything 3 (repackaged to ComfyUI's native Dinov2Model layout via scripts/convert_da3.py) + if '{}backbone.embeddings.patch_embeddings.projection.weight'.format(key_prefix) in state_dict_keys: + dit_config = {} + dit_config["image_model"] = "DepthAnything3" + + patch_w = state_dict['{}backbone.embeddings.patch_embeddings.projection.weight'.format(key_prefix)] + embed_dim = patch_w.shape[0] + depth = count_blocks(state_dict_keys, '{}backbone.encoder.layer.'.format(key_prefix) + '{}.') + + # Backbone preset is determined by embed_dim (matches vits/vitb/vitl/vitg). + backbone_name = {384: "vits", 768: "vitb", 1024: "vitl", 1536: "vitg"}.get(embed_dim) + if backbone_name is None: + return None + dit_config["backbone_name"] = backbone_name + + # Detect DA3 extensions on top of vanilla DINOv2. + has_camera_token = '{}backbone.embeddings.camera_token'.format(key_prefix) in state_dict_keys + # qk-norm shows up as `attention.q_norm.weight` on enabled blocks. + qknorm_indices = [ + i for i in range(depth) + if '{}backbone.encoder.layer.{}.attention.q_norm.weight'.format(key_prefix, i) in state_dict_keys + ] + qknorm_start = qknorm_indices[0] if qknorm_indices else -1 + + # The DA3 main-series configs always set alt_start == qknorm_start == rope_start. + # cat_token=True is implied by the presence of camera_token. + if has_camera_token: + dit_config["alt_start"] = qknorm_start + dit_config["rope_start"] = qknorm_start + dit_config["qknorm_start"] = qknorm_start + dit_config["cat_token"] = True + else: + dit_config["alt_start"] = -1 + dit_config["rope_start"] = -1 + dit_config["qknorm_start"] = -1 + dit_config["cat_token"] = False + + # Detect head type and config. + has_aux = '{}head.scratch.refinenet1_aux.out_conv.weight'.format(key_prefix) in state_dict_keys + dit_config["head_dim_in"] = state_dict['{}head.projects.0.weight'.format(key_prefix)].shape[1] + dit_config["head_features"] = state_dict['{}head.scratch.refinenet1.out_conv.weight'.format(key_prefix)].shape[0] + dit_config["head_out_channels"] = [ + state_dict['{}head.projects.{}.weight'.format(key_prefix, i)].shape[0] + for i in range(4) + ] + if has_aux: + # DualDPT: dim_in = 2 * embed_dim (because cat_token doubles token width). + dit_config["head_type"] = "dualdpt" + dit_config["head_output_dim"] = 2 + dit_config["head_use_sky_head"] = False + else: + dit_config["head_type"] = "dpt" + dit_config["head_output_dim"] = state_dict[ + '{}head.scratch.output_conv2.2.weight'.format(key_prefix) + ].shape[0] + dit_config["head_use_sky_head"] = ( + '{}head.scratch.sky_output_conv2.0.weight'.format(key_prefix) in state_dict_keys + ) + + # out_layers: hard-coded per upstream YAML config (depth-aware default). + if depth >= 24: + # vitl: depths used vary between DA3-Large (DualDPT) and Mono/Metric (DPT). + if has_aux: + dit_config["out_layers"] = [11, 15, 19, 23] + else: + dit_config["out_layers"] = [4, 11, 17, 23] + else: + # vits/vitb: 12 blocks + dit_config["out_layers"] = [5, 7, 9, 11] + + # Camera encoder/decoder presence (multi-view + pose path). + has_cam_enc = '{}cam_enc.token_norm.weight'.format(key_prefix) in state_dict_keys + has_cam_dec = '{}cam_dec.fc_t.weight'.format(key_prefix) in state_dict_keys + dit_config["has_cam_enc"] = has_cam_enc + dit_config["has_cam_dec"] = has_cam_dec + if has_cam_enc: + cam_enc_w = state_dict.get( + '{}cam_enc.pose_branch.fc2.weight'.format(key_prefix) + ) + if cam_enc_w is not None: + dit_config["cam_dim_out"] = cam_enc_w.shape[0] + if has_cam_dec: + cam_dec_w = state_dict.get( + '{}cam_dec.fc_t.weight'.format(key_prefix) + ) + if cam_dec_w is not None: + dit_config["cam_dec_dim_in"] = cam_dec_w.shape[1] + return dit_config + if '{}layers.0.mlp.linear_fc2.weight'.format(key_prefix) in state_dict_keys: # Ernie Image dit_config = {} dit_config["image_model"] = "ernie" diff --git a/comfy/supported_models.py b/comfy/supported_models.py index 42325d71c..3be935577 100644 --- a/comfy/supported_models.py +++ b/comfy/supported_models.py @@ -2056,6 +2056,23 @@ class RT_DETR_v4(supported_models_base.BASE): return None +class DepthAnything3(supported_models_base.BASE): + unet_config = { + "image_model": "DepthAnything3", + } + + # Mono path: no num_heads / num_head_channels needed. + unet_extra_config = {} + + supported_inference_dtypes = [torch.float16, torch.bfloat16, torch.float32] + + def get_model(self, state_dict, prefix="", device=None): + return model_base.DepthAnything3(self, device=device) + + def clip_target(self, state_dict={}): + return None + + class ErnieImage(supported_models_base.BASE): unet_config = { "image_model": "ernie", @@ -2298,4 +2315,5 @@ models = [ CogVideoX_I2V, CogVideoX_T2V, SVD_img2vid, + DepthAnything3, ] diff --git a/comfy_extras/nodes_depth_anything_3.py b/comfy_extras/nodes_depth_anything_3.py new file mode 100644 index 000000000..020112515 --- /dev/null +++ b/comfy_extras/nodes_depth_anything_3.py @@ -0,0 +1,681 @@ +"""ComfyUI nodes for Depth Anything 3. +Model capability matrix: + +Variant head_type has_sky has_conf cam_dec +DA3-Small dualdpt False True yes +DA3-Base dualdpt False True yes +DA3-Mono-Large dpt True False no +DA3-Metric-Large dpt True False no (raw output is metres) +""" + +from __future__ import annotations + +import logging +from typing_extensions import override + +import torch + +import comfy.model_management as mm +import comfy.sd +import folder_paths +from comfy.ldm.colormap import turbo as _turbo +from comfy.ldm.depth_anything_3 import preprocess as da3_preprocess +from comfy_api.latest import ComfyExtension, Types, io +from comfy.ldm.moge.geometry import triangulate_grid_mesh + +DA3ModelType = io.Custom("DA3_MODEL") +DA3Geometry = io.Custom("DA3_GEOMETRY") +DA3PointCloud = io.Custom("DA3_POINT_CLOUD") + +# DA3_GEOMETRY is a dict with these optional keys (absent when the upstream model didn't produce them): +# +# Per-frame tensors - B = batch size in mono mode; B = S (number of views) in multi-view mode. +# "depth": torch.Tensor (B, H, W) -- raw model depth (always present; matches MoGe convention) +# "image": torch.Tensor (B, H, W, 3) -- source image in [0, 1], CPU (always present) +# "mode": str -- "mono" or "multiview" (always present) +# "sky": torch.Tensor (B, H, W) -- sky probability in [0, 1] (Mono/Metric variants only) +# "confidence": torch.Tensor (B, H, W) -- raw model confidence output (Small/Base variants only) +# +# Multi-view only - S = number of views; the leading 1 is the scene dimension from the model. +# "extrinsics": torch.Tensor (1, S, 3, 4) -- world-to-camera [R|t] matrices +# "intrinsics": torch.Tensor (1, S, 3, 3) -- pixel-space intrinsics +# +# DA3_POINT_CLOUD is a dict: +# "points": torch.Tensor (N, 3) -- 3-D coords in glTF convention (Y-up, Z-back) +# "colors": torch.Tensor (N, 3) -- RGB in [0, 1], or None +# "confidence": torch.Tensor (N,) -- raw confidence per point, or None + + +def _da3_unproject(depth: torch.Tensor, K: torch.Tensor) -> torch.Tensor: + """Pixel-space K⁻¹ unprojection: (H,W) depth → (H,W,3) point map in OpenCV space.""" + H, W = depth.shape + u = torch.arange(W, dtype=torch.float32, device=depth.device) + v = torch.arange(H, dtype=torch.float32, device=depth.device) + u, v = torch.meshgrid(u, v, indexing='xy') # both (H, W) + pix = torch.stack([u, v, torch.ones_like(u)], dim=-1) # (H, W, 3) + rays = torch.einsum('ij,hwj->hwi', torch.linalg.inv(K.to(depth.device)), pix) + return rays * depth.unsqueeze(-1) # (H, W, 3) + + +def _da3_default_K(H: int, W: int) -> torch.Tensor: + """Fallback ~60° FOV pinhole K for mono-mode DA3 (no intrinsics in geometry).""" + fx = fy = float(W) * 0.7 + return torch.tensor([[fx, 0.0, (W - 1) / 2.0], + [0.0, fy, (H - 1) / 2.0], + [0.0, 0.0, 1.0]], dtype=torch.float32) + + +def _da3_get_K(geometry: dict, b: int, H: int, W: int) -> torch.Tensor: + """Return pixel-space K for batch element b, falling back to a default estimate.""" + if "intrinsics" in geometry: + # shape (1, S, 3, 3) - leading scene dimension from the multiview head + return geometry["intrinsics"][0, b].float() + logging.getLogger("comfy").warning( + "DA3_GEOMETRY has no intrinsics (mono-mode model). " + "Using a ~60° FOV estimate; 3-D reconstruction may be inaccurate." + ) + return _da3_default_K(H, W) + + +def _da3_get_extrinsic(geometry: dict, b: int) -> torch.Tensor | None: + """Return the world-to-camera extrinsic for batch element b, or None in mono mode. + + The model outputs (1, S, 3, 4) [R|t] matrices; the fallback identity is (4, 4). + _da3_apply_extrinsic handles both shapes via [:3, :3] / [:3, 3] slicing. + """ + if "extrinsics" not in geometry: + return None + return geometry["extrinsics"][0, b].float() + + +def _da3_apply_extrinsic(points_cam: torch.Tensor, E: torch.Tensor) -> torch.Tensor: + """Transform (H,W,3) OpenCV camera-space points to world space.""" + E = E.to(points_cam.device).float() + if not torch.isfinite(E).all(): + logging.getLogger("comfy").warning( + "DA3 extrinsic matrix contains non-finite values (pose estimation may have failed). " + "Falling back to camera-space coordinates." + ) + return points_cam + H, W, _ = points_cam.shape + R = E[:3, :3] # (3, 3) rotation + t = E[:3, 3] # (3,) translation + R_inv = R.T # rotation inverse = transpose for orthogonal R + t_inv = -(R_inv @ t) # (3,) + pts = points_cam.reshape(-1, 3) # (N, 3) + pts_world = pts @ R_inv.T + t_inv # (N, 3) + return pts_world.reshape(H, W, 3) + + +def _normalize_confidence(conf: torch.Tensor) -> torch.Tensor: + """Map raw confidence to [0, 1] per image.""" + B = conf.shape[0] + out = [] + for i in range(B): + c = conf[i] + c_min, c_max = c.min(), c.max() + out.append((c - c_min) / (c_max - c_min) if c_max > c_min else torch.ones_like(c)) + return torch.stack(out, dim=0) + + +def _da3_build_mask(geometry: dict, b: int, H: int, W: int, confidence_threshold: float, use_sky_mask: bool) -> torch.Tensor: + """Build (H,W) bool keep-mask from sky probability and confidence.""" + mask = torch.ones(H, W, dtype=torch.bool) + if use_sky_mask and "sky" in geometry: + mask = mask & (geometry["sky"][b] < 0.5) + if "confidence" in geometry and confidence_threshold > 0.0: + conf_norm = _normalize_confidence(geometry["confidence"][b:b + 1])[0] + mask = mask & (conf_norm >= confidence_threshold) + return mask + + +class LoadDA3Model(io.ComfyNode): + @classmethod + def define_schema(cls): + return io.Schema( + node_id="LoadDA3Model", + display_name="Load Depth Anything 3", + category="model/loaders", + inputs=[ + io.Combo.Input( + "model_name", + options=folder_paths.get_filename_list("geometry_estimation"), + ), + io.Combo.Input( + "weight_dtype", + options=["default", "fp16", "bf16", "fp32"], + default="default", + ), + ], + outputs=[DA3ModelType.Output()], + ) + + @classmethod + def execute(cls, model_name, weight_dtype) -> io.NodeOutput: + model_options = {} + if weight_dtype == "fp16": + model_options["dtype"] = torch.float16 + elif weight_dtype == "bf16": + model_options["dtype"] = torch.bfloat16 + elif weight_dtype == "fp32": + model_options["dtype"] = torch.float32 + + path = folder_paths.get_full_path_or_raise("geometry_estimation", model_name) + model = comfy.sd.load_diffusion_model(path, model_options=model_options) + return io.NodeOutput(model) + + +def _run_da3(model_patcher, image: torch.Tensor, process_res: int, method: str = "upper_bound_resize"): + """Run DA3 on (B,H,W,3), returns depth/conf/sky at original resolution (or None).""" + assert image.ndim == 4 and image.shape[-1] == 3, f"expected (B,H,W,3) IMAGE; got {tuple(image.shape)}" + + B, H, W, _ = image.shape + mm.load_model_gpu(model_patcher) + diffusion = model_patcher.model.diffusion_model + device = mm.get_torch_device() + dtype = diffusion.dtype if diffusion.dtype is not None else torch.float32 + + depths, confs, skies = [], [], [] + for i in range(B): + single = image[i:i + 1].to(device) + x = da3_preprocess.preprocess_image(single, process_res=process_res, method=method) + x = x.to(dtype=dtype) + with torch.no_grad(): + out = diffusion(x) + + depth_lr = out["depth"] + depth_full = torch.nn.functional.interpolate( + depth_lr.unsqueeze(1).float(), size=(H, W), + mode="bilinear", align_corners=False, + ).squeeze(1).cpu() + depths.append(depth_full) + + if "depth_conf" in out: + conf_full = torch.nn.functional.interpolate( + out["depth_conf"].unsqueeze(1).float(), size=(H, W), + mode="bilinear", align_corners=False, + ).squeeze(1).cpu() + confs.append(conf_full) + if "sky" in out: + sky_full = torch.nn.functional.interpolate( + out["sky"].unsqueeze(1).float(), size=(H, W), + mode="bilinear", align_corners=False, + ).squeeze(1).cpu() + skies.append(sky_full) + + depth = torch.cat(depths, dim=0) + confidence = torch.cat(confs, dim=0) if confs else None + sky = torch.cat(skies, dim=0) if skies else None + return depth, confidence, sky + + +class DA3Inference(io.ComfyNode): + @classmethod + def define_schema(cls): + return io.Schema( + node_id="DA3Inference", + search_aliases=["depth", "geometry", "da3", "depth anything", "monocular", "pointmap", "sky", "3d", "metric depth", "disparity"], + display_name="Run Depth Anything 3", + category="image/geometry estimation", + description="Run Depth Anything 3 on an image. In multi-view mode each image is treated as a separate view of the same scene.", + inputs=[ + DA3ModelType.Input("da3_model"), + io.Image.Input("image"), + io.Int.Input("resolution", default=504, min=140, max=2520, step=14, + tooltip="Resolution the model runs at (longest side, multiple of 14).\n" + "Lower = faster / less VRAM.\n" + "Higher = more detail.\n" + "Output is upsampled back to the original size."), + io.Combo.Input("resize_method", options=["upper_bound_resize", "lower_bound_resize"], default="upper_bound_resize", + tooltip="upper_bound_resize: scale so the longest side = resolution (caps memory, default).\n" + "lower_bound_resize: scale so the shortest side = resolution (preserves more detail on tall/wide images, uses more memory)."), + io.DynamicCombo.Input("mode", tooltip="mono: single view image (works with any model variant).\n" + "multiview: all images processed together for geometric consistency + camera pose (for Small/Base models only).", + options=[ + io.DynamicCombo.Option("mono", []), + io.DynamicCombo.Option("multiview", [ + io.Combo.Input("ref_view_strategy", options=["saddle_balanced", "saddle_sim_range", "first", "middle"], default="saddle_balanced", + tooltip="Which view acts as the geometric anchor.\n" + "- saddle_balanced: the view most 'average' across all others (best general choice).\n" + "- saddle_sim_range: the view most visually distinct from the others.\n" + "- first / middle: fixed positional picks."), + io.Combo.Input("pose_method", options=["cam_dec", "ray_pose"], default="cam_dec", + tooltip="How the camera field-of-view is estimated (for Small/Base models only).\n" + "- cam_dec: learned from image features.\n" + "- ray_pose: derived geometrically from the model's 3D ray output.\n" + "Affects perspective correctness of the 3D output. Try both if results look distorted."), + ]), + ]), + ], + outputs=[ + DA3Geometry.Output("da3_geometry", tooltip="Dictionary of non-normalized tensors.\n" + "Always has the keys: depth, image, mode.\n" + "Optional keys: sky (for Mono/Metric), confidence (for Small/Base), extrinsics + intrinsics (for multi-view)."), + ], + ) + + @classmethod + def execute(cls, da3_model, image, resolution, resize_method, mode) -> io.NodeOutput: + mode_val = mode["mode"] # "mono" or "multiview" + + if mode_val == "mono": + return cls._execute_mono(da3_model, image, resolution, resize_method) + + # Capability checks for multi-view mode. + diffusion = da3_model.model.diffusion_model + pose_method = mode["pose_method"] + ref_view_strategy = mode["ref_view_strategy"] + + has_cam_dec = diffusion.cam_dec is not None + has_dualdpt = diffusion.head_type == "dualdpt" + + if not has_cam_dec and not has_dualdpt: + raise ValueError( + "multi-view mode requires Small or Base model. The loaded model " + f"(head_type='{diffusion.head_type}') does not support cross-view " + "attention or camera pose estimation. Switch mode to 'mono', or " + "load Small or Base model for mult-view." + ) + + if pose_method == "cam_dec" and not has_cam_dec: + raise ValueError( + "pose_method='cam_dec' requires a camera decoder, but the loaded " + f"model (head_type='{diffusion.head_type}') does not have one. " + "Use pose_method='ray_pose' instead." + ) + if pose_method == "ray_pose" and not has_dualdpt: + raise ValueError( + "pose_method='ray_pose' requires a DualDPT head, but the loaded " + f"model has a '{diffusion.head_type}' head. " + "Use pose_method='cam_dec' instead." + ) + + return cls._execute_multiview( + da3_model, image, resolution, resize_method, + ref_view_strategy, pose_method, + ) + + @classmethod + def _execute_mono(cls, model, image, resolution, resize_method) -> io.NodeOutput: + depth, confidence, sky = _run_da3(model, image, resolution, method=resize_method) + + geometry: dict = { + "depth": depth.contiguous(), + "image": image[..., :3].cpu(), + "mode": "mono", + } + if sky is not None: + geometry["sky"] = sky.contiguous() + if confidence is not None: + geometry["confidence"] = confidence.contiguous() + return io.NodeOutput(geometry) + + @classmethod + def _execute_multiview(cls, model, image, resolution, resize_method, ref_view_strategy, pose_method) -> io.NodeOutput: + assert image.ndim == 4 and image.shape[-1] == 3, \ + f"expected (B,H,W,3) IMAGE; got {tuple(image.shape)}" + S, H, W, _ = image.shape + + mm.load_model_gpu(model) + diffusion = model.model.diffusion_model + device = mm.get_torch_device() + dtype = diffusion.dtype if diffusion.dtype is not None else torch.float32 + + # All views in a single forward pass: (1, S, 3, H', W'). + x = image.to(device) + x = da3_preprocess.preprocess_image(x, process_res=resolution, method=resize_method) + x = x.to(dtype=dtype).unsqueeze(0) + + use_ray_pose = (pose_method == "ray_pose") + with torch.no_grad(): + out = diffusion(x, use_ray_pose=use_ray_pose, ref_view_strategy=ref_view_strategy) + + depth = torch.nn.functional.interpolate( + out["depth"].float().unsqueeze(1), size=(H, W), + mode="bilinear", align_corners=False, + ).squeeze(1).cpu() + + sky = None + if "sky" in out: + sky = torch.nn.functional.interpolate( + out["sky"].unsqueeze(1).float(), size=(H, W), + mode="bilinear", align_corners=False, + ).squeeze(1).cpu() + + if "extrinsics" in out and "intrinsics" in out: + extrinsics = out["extrinsics"].float().cpu() + intrinsics = out["intrinsics"].float().cpu() + else: + extrinsics = torch.eye(4)[None, None].expand(1, S, 4, 4).clone() + intrinsics = torch.eye(3)[None, None].expand(1, S, 3, 3).clone() + + geometry: dict = { + "depth": depth.contiguous(), + "image": image[..., :3].cpu(), + "mode": "multiview", + "extrinsics": extrinsics.contiguous(), + "intrinsics": intrinsics.contiguous(), + } + if sky is not None: + geometry["sky"] = sky.contiguous() + if "depth_conf" in out: + conf = torch.nn.functional.interpolate( + out["depth_conf"].unsqueeze(1).float(), size=(H, W), + mode="bilinear", align_corners=False, + ).squeeze(1).cpu() + geometry["confidence"] = conf.contiguous() + return io.NodeOutput(geometry) + + +class DA3Render(io.ComfyNode): + """Render a visualization from a DA3_GEOMETRY packet.""" + + _DEPTH_RENDER_INPUTS = [ + io.Combo.Input("normalization", + options=["v2_style", "min_max", "raw"], + default="v2_style", + tooltip="- v2_style: mean/std normalisation for perceptually balanced results (default).\n" + "- min_max: stretches the full depth range to [0, 1] for maximum contrast.\n" + "- raw: no scaling,preserves metric units for Metric model."), + io.Boolean.Input("apply_sky_clip", default=False, + tooltip="Clip sky-region depth to the 99th percentile of foreground depth before normalisation. " + "Requires a sky key in the da3_geometry input (for Mono/Metric models only)."), + ] + + @classmethod + def define_schema(cls): + return io.Schema( + node_id="DA3Render", + display_name="Render Depth Anything 3", + category="image/geometry estimation", + description="Render a depth map, confidence map, or sky mask from Depth Anything 3 geometry data.", + inputs=[ + DA3Geometry.Input("da3_geometry"), + io.DynamicCombo.Input("output", + tooltip="- depth: normalised greyscale depth image.\n" + "- depth_colored: depth mapped through the Turbo colormap.\n" + "- sky_mask: sky probability in [0, 1] (for Mono/Metric models only).\n" + "- confidence: normalised depth confidence (for Small/Base models only).", + options=[ + io.DynamicCombo.Option("depth", cls._DEPTH_RENDER_INPUTS), + io.DynamicCombo.Option("depth_colored", cls._DEPTH_RENDER_INPUTS), + io.DynamicCombo.Option("sky_mask", [ + io.Boolean.Input("colored", default=False, tooltip="Apply the Turbo colormap to the sky mask."), + ]), + io.DynamicCombo.Option("confidence", [ + io.Boolean.Input("colored", default=False, tooltip="Apply the Turbo colormap to the confidence map."), + ]), + ]), + ], + outputs=[io.Image.Output()], + ) + + @classmethod + def execute(cls, da3_geometry, output) -> io.NodeOutput: + output_val = output["output"] + + if output_val in ("depth", "depth_colored"): + normalization = output["normalization"] + apply_sky_clip = output["apply_sky_clip"] + if apply_sky_clip and "sky" not in da3_geometry: + raise ValueError( + "apply_sky_clip=True requires a sky tensor in the da3_geometry input, but none is present. " + "Run with Mono/Metric models or set apply_sky_clip=False." + ) + depth = da3_geometry["depth"] + sky = da3_geometry.get("sky") + if apply_sky_clip and sky is not None: + depth = torch.stack([ + da3_preprocess.apply_sky_aware_clip(depth[i], sky[i]) + for i in range(depth.shape[0]) + ], dim=0) + grey = cls._depth_to_image(depth, sky, normalization) # (B,H,W,3) greyscale + result = _turbo(grey[..., 0]) if output_val == "depth_colored" else grey + + elif output_val == "sky_mask": + if "sky" not in da3_geometry: + raise ValueError("geometry has no sky output; run with Mono/Metric models.") + sky = da3_geometry["sky"] + if output["colored"]: + result = _turbo(sky) + else: + result = sky.unsqueeze(-1).expand(*sky.shape, 3).contiguous() + + elif output_val == "confidence": + if "confidence" not in da3_geometry: + raise ValueError("da3_geometry has no confidence output; run with Small/Base models.") + conf = _normalize_confidence(da3_geometry["confidence"]) + if output["colored"]: + result = _turbo(conf) + else: + result = conf.unsqueeze(-1).expand(*conf.shape, 3).contiguous() + + else: + raise ValueError(f"Unknown output mode: {output_val}") + + return io.NodeOutput(result.float()) + + @staticmethod + def _depth_to_image(depth: torch.Tensor, sky_for_norm: torch.Tensor | None, normalization: str) -> torch.Tensor: + """Normalise depth and pack as an (B,H,W,3) image tensor.""" + + N = depth.shape[0] + if normalization == "v2_style": + norm = torch.stack([ + da3_preprocess.normalize_depth_v2_style( + depth[i], sky_for_norm[i] if sky_for_norm is not None else None) + for i in range(N) + ], dim=0) + elif normalization == "min_max": + norm = da3_preprocess.normalize_depth_min_max(depth) + else: + norm = depth + + out = norm.unsqueeze(-1).repeat(1, 1, 1, 3) + if normalization != "raw": + out = out.clamp(0.0, 1.0) + return out.contiguous() + + +class DA3GeometryToMesh(io.ComfyNode): + """Convert a DA3_GEOMETRY packet into a Types.MESH by unprojecting depth and triangulating.""" + + @classmethod + def define_schema(cls): + return io.Schema( + node_id="DA3GeometryToMesh", + search_aliases=["da3", "depth anything", "mesh", "geometry", "3d", "triangulate"], + display_name="Convert DA3 Geometry to Mesh", + category="image/geometry estimation", + description="Convert a depth map into a triangulated 3D mesh.", + inputs=[ + DA3Geometry.Input("da3_geometry"), + io.Int.Input("batch_index", default=0, min=0, max=4096, tooltip="Which image of a batch to convert. Per-image vertex counts differ so batches cannot be stacked."), + io.Int.Input("decimation", default=1, min=1, max=8, tooltip="Vertex stride. 1 = full resolution, 2 = half, etc."), + io.Float.Input("discontinuity_threshold", default=0.04, min=0.0, max=1.0, step=0.01, tooltip="Drop triangles whose 3x3 depth span exceeds this fraction. 0 = off."), + io.Float.Input("confidence_threshold", default=0.1, min=0.0, max=1.0, step=0.01, + tooltip="Exclude pixels whose per-image normalised confidence is below this value (0 = keep all, 1 = keep only the single most confident pixel). " + "Used when the geometry has a confidence map (Small/Base models)."), + io.Boolean.Input("use_sky_mask", default=True, tooltip="Exclude sky-probability pixels (sky >= 0.5) from the mesh. Used when the geometry has a sky map (Mono/Metric models)."), + io.Boolean.Input("texture", default=True, tooltip="Use the source image as a base color texture."), + ], + outputs=[io.Mesh.Output()], + ) + + @classmethod + def execute(cls, da3_geometry, batch_index, decimation, discontinuity_threshold, confidence_threshold, use_sky_mask, texture) -> io.NodeOutput: + depth_all = da3_geometry["depth"] # (B, H, W) + B = depth_all.shape[0] + if batch_index >= B: + raise ValueError(f"batch_index {batch_index} is out of range; DA3_GEOMETRY has batch size {B}.") + + depth = depth_all[batch_index] # (H, W) + H, W = depth.shape + + # NaN/inf depth would propagate silently through unproject and produce an + # empty mesh; replace them with 0 here so those pixels are later excluded + # by the isfinite check inside triangulate_grid_mesh. + depth = depth.clone() + n_bad = (~torch.isfinite(depth)).sum().item() + if n_bad: + logging.getLogger("comfy").warning( + f"DA3GeometryToMesh: depth[{batch_index}] has {n_bad} non-finite pixels " + f"({100*n_bad/(H*W):.1f}%) - zeroed before unproject." + ) + depth[~torch.isfinite(depth)] = 0.0 + logging.getLogger("comfy").debug( + f"DA3GeometryToMesh: depth[{batch_index}] range " + f"[{depth.min():.4g}, {depth.max():.4g}], mean={depth.mean():.4g}" + ) + + K = _da3_get_K(da3_geometry, batch_index, H, W) + points = _da3_unproject(depth, K) # (H, W, 3) in OpenCV camera space + + # Apply world-to-camera inverse so multi-view frames share a common world frame. + E = _da3_get_extrinsic(da3_geometry, batch_index) + if E is not None: + points = _da3_apply_extrinsic(points, E) + + # Mask invalid pixels by setting them to inf so triangulate_grid_mesh skips them. + mask = _da3_build_mask(da3_geometry, batch_index, H, W, confidence_threshold, use_sky_mask) + # Also exclude pixels where depth was invalid. + mask = mask & (depth_all[batch_index] > 0) & torch.isfinite(depth_all[batch_index]) + points = points.clone() + points[~mask] = float('inf') + + verts, faces, uvs = triangulate_grid_mesh( + points, + decimation=decimation, + discontinuity_threshold=discontinuity_threshold, + depth=depth, + ) + if verts.shape[0] == 0 or faces.shape[0] == 0: + raise ValueError( + "DA3GeometryToMesh produced an empty mesh. " + "Try raising discontinuity_threshold, lowering confidence_threshold, " + "or disabling use_sky_mask." + ) + + # OpenCV (X right, Y down, Z forward) → glTF (X right, Y up, Z back). + # Same transform as MoGePointMapToMesh perspective branch. + verts = verts * torch.tensor([1.0, -1.0, -1.0], dtype=verts.dtype) + faces = faces[:, [0, 2, 1]].contiguous() + + tex = da3_geometry["image"][batch_index:batch_index + 1] if texture else None + mesh = Types.MESH( + vertices=verts.unsqueeze(0), + faces=faces.unsqueeze(0), + uvs=uvs.unsqueeze(0), + texture=tex, + ) + return io.NodeOutput(mesh) + + +class DA3GeometryToPointCloud(io.ComfyNode): + """Unproject a DA3_GEOMETRY depth map into a filtered DA3_POINT_CLOUD.""" + + @classmethod + def define_schema(cls): + return io.Schema( + node_id="DA3GeometryToPointCloud", + search_aliases=["da3", "depth anything", "point cloud", "pointcloud", "3d", "geometry"], + display_name="Convert DA3 Geometry to Point Cloud", + category="image/geometry estimation", + description="Convert a depth map into a 3D point cloud.", + inputs=[ + DA3Geometry.Input("da3_geometry"), + io.Int.Input("batch_index", default=0, min=0, max=4096, tooltip="Which image of a batch to convert."), + io.Float.Input("confidence_threshold", default=0.1, min=0.0, max=1.0, step=0.01, + tooltip="Exclude pixels whose per-image normalised confidence is below this value (0 = keep all). Used when the geometry has a confidence map (Small/Base models)."), + io.Boolean.Input("use_sky_mask", default=True, + tooltip="Exclude sky-probability pixels (sky >= 0.5). Used when the geometry has a sky map (Mono/Metric models)."), + io.Int.Input("downsample", default=1, min=1, max=16, + tooltip="Take every Nth pixel (1 = full resolution). Higher values give fewer points and faster processing."), + ], + # TODO: add a proper PointCloud output type + outputs=[DA3PointCloud.Output(display_name="point_cloud")], + ) + + @classmethod + def execute(cls, da3_geometry, batch_index, confidence_threshold, use_sky_mask, downsample) -> io.NodeOutput: + depth_all = da3_geometry["depth"] # (B, H, W) + B = depth_all.shape[0] + if batch_index >= B: + raise ValueError(f"batch_index {batch_index} is out of range; DA3_GEOMETRY has batch size {B}.") + + depth = depth_all[batch_index].clone() # (H, W) + depth[~torch.isfinite(depth)] = 0.0 + H, W = depth.shape + + K = _da3_get_K(da3_geometry, batch_index, H, W) + + if downsample > 1: + depth = depth[::downsample, ::downsample].contiguous() + # Scale intrinsics to the downsampled grid. + K = K.clone() + K[0, :] /= downsample + K[1, :] /= downsample + + H_ds, W_ds = depth.shape + points = _da3_unproject(depth, K) # (H_ds, W_ds, 3) in OpenCV camera space + + # Apply world-to-camera inverse so multi-view frames share a common world frame. + E = _da3_get_extrinsic(da3_geometry, batch_index) + if E is not None: + points = _da3_apply_extrinsic(points, E) + + # Rebuild mask at downsampled resolution. + mask = _da3_build_mask(da3_geometry, batch_index, H, W, confidence_threshold, use_sky_mask) + if downsample > 1: + mask = mask[::downsample, ::downsample] + + mask = mask & torch.isfinite(depth) + + # OpenCV → glTF: flip Y and Z. + points_gltf = points.clone() + points_gltf[..., 1] *= -1.0 + points_gltf[..., 2] *= -1.0 + + pts_flat = points_gltf.reshape(-1, 3)[mask.reshape(-1)] + + colors_flat = None + if "image" in da3_geometry: + img = da3_geometry["image"][batch_index] # (H, W, 3) + if downsample > 1: + img = img[::downsample, ::downsample] + colors_flat = img.reshape(-1, 3)[mask.reshape(-1)] + + conf_flat = None + if "confidence" in da3_geometry: + conf = da3_geometry["confidence"][batch_index] # (H, W) + if downsample > 1: + conf = conf[::downsample, ::downsample] + conf_flat = conf.reshape(-1)[mask.reshape(-1)] + + if pts_flat.shape[0] == 0: + raise ValueError( + "DA3GeometryToPointCloud produced zero points after filtering. " + "Try lowering confidence_threshold or disabling use_sky_mask." + ) + + return io.NodeOutput({ + "points": pts_flat, + "colors": colors_flat, + "confidence": conf_flat, + }) + + +class DA3Extension(ComfyExtension): + @override + async def get_node_list(self) -> list[type[io.ComfyNode]]: + return [ + LoadDA3Model, + DA3Inference, + DA3Render, + DA3GeometryToMesh, + # DA3GeometryToPointCloud, # Keep this commented out for now until we have a proper PointCloud output type + ] + + +async def comfy_entrypoint() -> DA3Extension: + return DA3Extension() diff --git a/comfy_extras/nodes_moge.py b/comfy_extras/nodes_moge.py index 422949531..a63f0414b 100644 --- a/comfy_extras/nodes_moge.py +++ b/comfy_extras/nodes_moge.py @@ -8,6 +8,7 @@ import folder_paths from comfy_api.latest import ComfyExtension, Types, io from typing_extensions import override +from comfy.ldm.colormap import turbo as _turbo from comfy.ldm.moge.model import MoGeModel from comfy.ldm.moge.geometry import triangulate_grid_mesh from comfy.ldm.moge.panorama import get_panorama_cameras, split_panorama_image, merge_panorama_depth, spherical_uv_to_directions, _uv_grid @@ -27,19 +28,6 @@ MoGeGeometry = io.Custom("MOGE_GEOMETRY") # "image": torch.Tensor (B, H, W, 3) in [0, 1], CPU (always present) -def _turbo(x: torch.Tensor) -> torch.Tensor: - """Anton Mikhailov polynomial approximation of the turbo colormap.""" - x = x.clamp(0.0, 1.0) - x2 = x * x - x3 = x2 * x - x4 = x2 * x2 - x5 = x4 * x - r = 0.13572138 + 4.61539260*x - 42.66032258*x2 + 132.13108234*x3 - 152.94239396*x4 + 59.28637943*x5 - g = 0.09140261 + 2.19418839*x + 4.84296658*x2 - 14.18503333*x3 + 4.27729857*x4 + 2.82956604*x5 - b = 0.10667330 + 12.64194608*x - 60.58204836*x2 + 110.36276771*x3 - 89.90310912*x4 + 27.34824973*x5 - return torch.stack([r, g, b], dim=-1).clamp(0.0, 1.0) - - def _normals_from_points(points: torch.Tensor) -> torch.Tensor: """Camera-space surface normals from a (B, H, W, 3) point map (v1 fallback).""" finite = torch.isfinite(points).all(dim=-1) diff --git a/nodes.py b/nodes.py index fb6952bad..0d422d418 100644 --- a/nodes.py +++ b/nodes.py @@ -2459,7 +2459,8 @@ async def init_builtin_extra_nodes(): "nodes_moge.py", "nodes_mediapipe.py", "nodes_gaussian_splat.py", - "nodes_triposplat.py" + "nodes_triposplat.py", + "nodes_depth_anything_3.py", ] import_failed = [] From 5fcf7a4a0f7786912733f2ec8a6808ad614ea74e Mon Sep 17 00:00:00 2001 From: comfyanonymous <121283862+comfyanonymous@users.noreply.github.com> Date: Tue, 9 Jun 2026 18:39:24 -0700 Subject: [PATCH 05/21] Always enable cuda malloc on cu130 and higher. (#14381) --- cuda_malloc.py | 16 +++++++++++++++- 1 file changed, 15 insertions(+), 1 deletion(-) diff --git a/cuda_malloc.py b/cuda_malloc.py index f7651981c..8c4422db8 100644 --- a/cuda_malloc.py +++ b/cuda_malloc.py @@ -2,6 +2,7 @@ import os import importlib.util from comfy.cli_args import args, PerformanceFeature import subprocess +import re #Can't use pytorch to get the GPU names because the cuda malloc has to be set before the first import. def get_gpu_names(): @@ -77,11 +78,24 @@ try: except: pass +def get_raw_cuda_version(version_str): + match = re.search(r'\+cu(\d+)', version_str) + if match: + try: + return int(match.group(1)) + except: + pass + return None + if not args.cuda_malloc: try: if int(version[0]) >= 2 and "+cu" in version: # enable by default for torch version 2.0 and up only on cuda torch if PerformanceFeature.AutoTune not in args.fast: # Autotune has issues with cuda malloc - args.cuda_malloc = cuda_malloc_supported() + cuda_version = get_raw_cuda_version(version) + if cuda_version is not None and cuda_version >= 130: + args.cuda_malloc = True + else: + args.cuda_malloc = cuda_malloc_supported() except: pass From 46d45aade1dfab6d5a3658f2650a4626f175be3a Mon Sep 17 00:00:00 2001 From: Comfy Org PR Bot Date: Wed, 10 Jun 2026 10:58:42 +0900 Subject: [PATCH 06/21] chore(openapi): sync shared API contract from cloud@ca12913 (#14367) --- openapi.yaml | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/openapi.yaml b/openapi.yaml index 2510f97d0..c27ed7adf 100644 --- a/openapi.yaml +++ b/openapi.yaml @@ -1960,8 +1960,8 @@ paths: schema: properties: hash: - description: Hash of the existing asset. Supports Blake3 (blake3:) or SHA256 (sha256:) formats - pattern: ^(blake3|sha256):[a-f0-9]{64}$ + description: 'Blake3 content hash of the existing asset (blake3: prefix)' + pattern: ^blake3:[a-f0-9]{64}$ type: string mime_type: description: MIME type of the asset (e.g., "image/png", "video/mp4") From f350acdf213a1b3cbeab2059888265b21590ce9f Mon Sep 17 00:00:00 2001 From: Kohaku-Blueleaf <59680068+KohakuBlueleaf@users.noreply.github.com> Date: Wed, 10 Jun 2026 11:07:47 +0800 Subject: [PATCH 07/21] [Trainer/bug] Ensure model is not inference mode (CORE-72) (#13400) * Ensure model is not inference mode * force clone inside training mode to avoid inference tensor * Allow force deepcopy for model patcher --- comfy/model_patcher.py | 7 ++-- comfy_extras/nodes_train.py | 74 ++++++++++++++++++------------------- 2 files changed, 41 insertions(+), 40 deletions(-) diff --git a/comfy/model_patcher.py b/comfy/model_patcher.py index b716a69e2..d70b42bf8 100644 --- a/comfy/model_patcher.py +++ b/comfy/model_patcher.py @@ -379,10 +379,11 @@ class ModelPatcher: def get_clone_model_override(self): return self.model, (self.backup, self.backup_buffers, self.object_patches_backup, self.pinned) - def clone(self, disable_dynamic=False, model_override=None): + def clone(self, disable_dynamic=False, model_override=None, force_deepcopy=False): class_ = self.__class__ - if self.is_dynamic() and disable_dynamic: - class_ = ModelPatcher + if self.is_dynamic() and disable_dynamic or force_deepcopy: + if self.is_dynamic() and disable_dynamic: + class_ = ModelPatcher if model_override is None: if self.cached_patcher_init is None: raise RuntimeError("Cannot create non-dynamic delegate: cached_patcher_init is not initialized.") diff --git a/comfy_extras/nodes_train.py b/comfy_extras/nodes_train.py index 273f55e7c..bb68da6fa 100644 --- a/comfy_extras/nodes_train.py +++ b/comfy_extras/nodes_train.py @@ -1149,45 +1149,45 @@ class TrainLoraNode(io.ComfyNode): # Process conditioning positive = _process_conditioning(positive) - # Setup model and dtype - mp = model.clone() - use_grad_scaler = False - lora_dtype = node_helpers.string_to_torch_dtype(lora_dtype) - if training_dtype != "none": - dtype = node_helpers.string_to_torch_dtype(training_dtype) - mp.set_model_compute_dtype(dtype) - else: - # Detect model's native dtype for autocast - model_dtype = mp.model.get_dtype() - if model_dtype == torch.float16: - dtype = torch.float16 - # GradScaler only supports float16 gradients, not bfloat16. - # Only enable it when lora params will also be in float16. - if lora_dtype != torch.bfloat16: - use_grad_scaler = True - # Warn about fp16 accumulation instability during training - if PerformanceFeature.Fp16Accumulation in args.fast: - logging.warning( - "WARNING: FP16 model detected with fp16_accumulation enabled. " - "This combination can be numerically unstable during training and may cause NaN values. " - "Suggested fixes: 1) Set training_dtype to 'bf16', or 2) Disable fp16_accumulation (remove from --fast flags)." - ) - else: - # For fp8, bf16, or other dtypes, use bf16 autocast - dtype = torch.bfloat16 - - # Prepare latents and compute counts - latents_dtype = dtype if dtype not in (None,) else torch.bfloat16 - latents, num_images, multi_res = _prepare_latents_and_count( - latents, latents_dtype, bucket_mode - ) - - # Validate and expand conditioning - positive = _validate_and_expand_conditioning(positive, num_images, bucket_mode) - with torch.inference_mode(False): + # Setup model and dtype + mp = model.clone(force_deepcopy=True) + use_grad_scaler = False + lora_dtype = node_helpers.string_to_torch_dtype(lora_dtype) + if training_dtype != "none": + dtype = node_helpers.string_to_torch_dtype(training_dtype) + mp.set_model_compute_dtype(dtype) + else: + # Detect model's native dtype for autocast + model_dtype = mp.model.get_dtype() + if model_dtype == torch.float16: + dtype = torch.float16 + # GradScaler only supports float16 gradients, not bfloat16. + # Only enable it when lora params will also be in float16. + if lora_dtype != torch.bfloat16: + use_grad_scaler = True + # Warn about fp16 accumulation instability during training + if PerformanceFeature.Fp16Accumulation in args.fast: + logging.warning( + "WARNING: FP16 model detected with fp16_accumulation enabled. " + "This combination can be numerically unstable during training and may cause NaN values. " + "Suggested fixes: 1) Set training_dtype to 'bf16', or 2) Disable fp16_accumulation (remove from --fast flags)." + ) + else: + # For fp8, bf16, or other dtypes, use bf16 autocast + dtype = torch.bfloat16 + + # Prepare latents and compute counts + latents_dtype = dtype if dtype not in (None,) else torch.bfloat16 + latents, num_images, multi_res = _prepare_latents_and_count( + latents, latents_dtype, bucket_mode + ) + + # Validate and expand conditioning + positive = _validate_and_expand_conditioning(positive, num_images, bucket_mode) + # Setup models for training - mp.model.requires_grad_(False) + mp.model.requires_grad_(False).train() # Load existing LoRA weights if provided existing_weights, existing_steps = _load_existing_lora(existing_lora) From a76bb4380ee9fcc0fc96e5bc9fe25d66ad7ca412 Mon Sep 17 00:00:00 2001 From: Matt Miller Date: Tue, 9 Jun 2026 21:07:10 -0700 Subject: [PATCH 08/21] chore(assets): drop vestigial tags.tag_type column (#14248) MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit tag_type was always "user" in practice — no code path ever set it to anything else (no system/seeded classification was wired up) and nothing queried it. The column, its ix_tags_tag_type index, and the TagUsage.type API field were dead weight, so they're removed. Adds alembic migration 0004 to drop the column and index. Verified: asset-seeder tests pass; migration applies cleanly on a fresh SQLite (tags retains only name; tag_type column + index dropped). Co-authored-by: guill --- alembic_db/versions/0004_drop_tag_type.py | 39 +++++++++++++++++++ app/assets/api/routes.py | 4 +- app/assets/api/schemas_out.py | 1 - app/assets/database/models.py | 3 -- app/assets/database/queries/tags.py | 13 +++---- app/assets/scanner.py | 2 +- app/assets/services/schemas.py | 1 - app/assets/services/tagging.py | 2 +- tests-unit/assets_test/queries/test_tags.py | 31 ++++++--------- .../assets_test/services/test_tagging.py | 10 ++--- .../assets_test/test_sync_references.py | 2 +- 11 files changed, 66 insertions(+), 42 deletions(-) create mode 100644 alembic_db/versions/0004_drop_tag_type.py diff --git a/alembic_db/versions/0004_drop_tag_type.py b/alembic_db/versions/0004_drop_tag_type.py new file mode 100644 index 000000000..582bec4e8 --- /dev/null +++ b/alembic_db/versions/0004_drop_tag_type.py @@ -0,0 +1,39 @@ +""" +Drop the vestigial tags.tag_type column. + +tag_type was always "user" in practice — no code path ever set it to anything +else (no system/seeded classification was ever wired up) and nothing queried it. +The column, its index (ix_tags_tag_type), and the corresponding API field were +dead weight, so they are removed. + +Revision ID: 0004_drop_tag_type +Revises: 0003_add_metadata_job_id +Create Date: 2026-06-03 +""" + +from alembic import op +import sqlalchemy as sa + +revision = "0004_drop_tag_type" +down_revision = "0003_add_metadata_job_id" +branch_labels = None +depends_on = None + + +def upgrade() -> None: + with op.batch_alter_table("tags") as batch_op: + batch_op.drop_index("ix_tags_tag_type") + batch_op.drop_column("tag_type") + + +def downgrade() -> None: + with op.batch_alter_table("tags") as batch_op: + batch_op.add_column( + sa.Column( + "tag_type", + sa.String(length=32), + nullable=False, + server_default="user", + ) + ) + batch_op.create_index("ix_tags_tag_type", ["tag_type"]) diff --git a/app/assets/api/routes.py b/app/assets/api/routes.py index 6555974e9..252ddfe8f 100644 --- a/app/assets/api/routes.py +++ b/app/assets/api/routes.py @@ -575,8 +575,8 @@ async def get_tags(request: web.Request) -> web.Response: ) tags = [ - schemas_out.TagUsage(name=name, count=count, type=tag_type) - for (name, tag_type, count) in rows + schemas_out.TagUsage(name=name, count=count) + for (name, count) in rows ] payload = schemas_out.TagsList( tags=tags, total=total, has_more=(query.offset + len(tags)) < total diff --git a/app/assets/api/schemas_out.py b/app/assets/api/schemas_out.py index 0e748b907..143848329 100644 --- a/app/assets/api/schemas_out.py +++ b/app/assets/api/schemas_out.py @@ -46,7 +46,6 @@ class AssetsList(BaseModel): class TagUsage(BaseModel): name: str count: int - type: str class TagsList(BaseModel): diff --git a/app/assets/database/models.py b/app/assets/database/models.py index a3af8a192..9b61d309a 100644 --- a/app/assets/database/models.py +++ b/app/assets/database/models.py @@ -227,7 +227,6 @@ class Tag(Base): __tablename__ = "tags" name: Mapped[str] = mapped_column(String(512), primary_key=True) - tag_type: Mapped[str] = mapped_column(String(32), nullable=False, default="user") asset_reference_links: Mapped[list[AssetReferenceTag]] = relationship( back_populates="tag", @@ -240,7 +239,5 @@ class Tag(Base): overlaps="asset_reference_links,tag_links,tags,asset_reference", ) - __table_args__ = (Index("ix_tags_tag_type", "tag_type"),) - def __repr__(self) -> str: return f"" diff --git a/app/assets/database/queries/tags.py b/app/assets/database/queries/tags.py index f4126dba8..d41d73a10 100644 --- a/app/assets/database/queries/tags.py +++ b/app/assets/database/queries/tags.py @@ -55,13 +55,11 @@ def validate_tags_exist(session: Session, tags: list[str]) -> None: raise ValueError(f"Unknown tags: {missing}") -def ensure_tags_exist( - session: Session, names: Iterable[str], tag_type: str = "user" -) -> None: +def ensure_tags_exist(session: Session, names: Iterable[str]) -> None: wanted = normalize_tags(list(names)) if not wanted: return - rows = [{"name": n, "tag_type": tag_type} for n in list(dict.fromkeys(wanted))] + rows = [{"name": n} for n in list(dict.fromkeys(wanted))] ins = ( sqlite.insert(Tag) .values(rows) @@ -97,7 +95,7 @@ def set_reference_tags( to_remove = [t for t in current if t not in desired] if to_add: - ensure_tags_exist(session, to_add, tag_type="user") + ensure_tags_exist(session, to_add) session.add_all( [ AssetReferenceTag( @@ -142,7 +140,7 @@ def add_tags_to_reference( return AddTagsResult(added=[], already_present=[], total_tags=total) if create_if_missing: - ensure_tags_exist(session, norm, tag_type="user") + ensure_tags_exist(session, norm) current = set(get_reference_tags(session, reference_id)) @@ -289,7 +287,6 @@ def list_tags_with_usage( q = ( select( Tag.name, - Tag.tag_type, func.coalesce(counts_sq.c.cnt, 0).label("count"), ) .select_from(Tag) @@ -331,7 +328,7 @@ def list_tags_with_usage( rows = (session.execute(q.limit(limit).offset(offset))).all() total = (session.execute(total_q)).scalar_one() - rows_norm = [(name, ttype, int(count or 0)) for (name, ttype, count) in rows] + rows_norm = [(name, int(count or 0)) for (name, count) in rows] return rows_norm, int(total or 0) diff --git a/app/assets/scanner.py b/app/assets/scanner.py index 495c30443..2c1e97840 100644 --- a/app/assets/scanner.py +++ b/app/assets/scanner.py @@ -355,7 +355,7 @@ def insert_asset_specs(specs: list[SeedAssetSpec], tag_pool: set[str]) -> int: return 0 with create_session() as sess: if tag_pool: - ensure_tags_exist(sess, tag_pool, tag_type="user") + ensure_tags_exist(sess, tag_pool) result = batch_insert_seed_assets(sess, specs=specs, owner_id="") sess.commit() return result.inserted_refs diff --git a/app/assets/services/schemas.py b/app/assets/services/schemas.py index 0eb128f58..2a52d76ca 100644 --- a/app/assets/services/schemas.py +++ b/app/assets/services/schemas.py @@ -56,7 +56,6 @@ class IngestResult: class TagUsage(NamedTuple): name: str - tag_type: str count: int diff --git a/app/assets/services/tagging.py b/app/assets/services/tagging.py index 37b612753..5fa39d26a 100644 --- a/app/assets/services/tagging.py +++ b/app/assets/services/tagging.py @@ -75,7 +75,7 @@ def list_tags( owner_id=owner_id, ) - return [TagUsage(name, tag_type, count) for name, tag_type, count in rows], total + return [TagUsage(name, count) for name, count in rows], total def list_tag_histogram( diff --git a/tests-unit/assets_test/queries/test_tags.py b/tests-unit/assets_test/queries/test_tags.py index 4ed99aa37..6222714d1 100644 --- a/tests-unit/assets_test/queries/test_tags.py +++ b/tests-unit/assets_test/queries/test_tags.py @@ -40,15 +40,15 @@ def _make_reference(session: Session, asset: Asset, name: str = "test", owner_id class TestEnsureTagsExist: def test_creates_new_tags(self, session: Session): - ensure_tags_exist(session, ["alpha", "beta"], tag_type="user") + ensure_tags_exist(session, ["alpha", "beta"]) session.commit() tags = session.query(Tag).all() assert {t.name for t in tags} == {"alpha", "beta"} def test_is_idempotent(self, session: Session): - ensure_tags_exist(session, ["alpha"], tag_type="user") - ensure_tags_exist(session, ["alpha"], tag_type="user") + ensure_tags_exist(session, ["alpha"]) + ensure_tags_exist(session, ["alpha"]) session.commit() assert session.query(Tag).count() == 1 @@ -65,13 +65,6 @@ class TestEnsureTagsExist: session.commit() assert session.query(Tag).count() == 0 - def test_tag_type_is_set(self, session: Session): - ensure_tags_exist(session, ["system-tag"], tag_type="system") - session.commit() - - tag = session.query(Tag).filter_by(name="system-tag").one() - assert tag.tag_type == "system" - class TestGetReferenceTags: def test_returns_empty_for_no_tags(self, session: Session): @@ -193,7 +186,7 @@ class TestMissingTagFunctions: def test_add_missing_tag_for_asset_id(self, session: Session): asset = _make_asset(session, "hash1") ref = _make_reference(session, asset) - ensure_tags_exist(session, ["missing"], tag_type="system") + ensure_tags_exist(session, ["missing"]) add_missing_tag_for_asset_id(session, asset_id=asset.id) session.commit() @@ -204,7 +197,7 @@ class TestMissingTagFunctions: def test_add_missing_tag_is_idempotent(self, session: Session): asset = _make_asset(session, "hash1") ref = _make_reference(session, asset) - ensure_tags_exist(session, ["missing"], tag_type="system") + ensure_tags_exist(session, ["missing"]) add_missing_tag_for_asset_id(session, asset_id=asset.id) add_missing_tag_for_asset_id(session, asset_id=asset.id) @@ -216,7 +209,7 @@ class TestMissingTagFunctions: def test_remove_missing_tag_for_asset_id(self, session: Session): asset = _make_asset(session, "hash1") ref = _make_reference(session, asset) - ensure_tags_exist(session, ["missing"], tag_type="system") + ensure_tags_exist(session, ["missing"]) add_missing_tag_for_asset_id(session, asset_id=asset.id) remove_missing_tag_for_asset_id(session, asset_id=asset.id) @@ -237,7 +230,7 @@ class TestListTagsWithUsage: rows, total = list_tags_with_usage(session) - tag_dict = {name: count for name, _, count in rows} + tag_dict = {name: count for name, count in rows} assert tag_dict["used"] == 1 assert tag_dict["unused"] == 0 assert total == 2 @@ -252,7 +245,7 @@ class TestListTagsWithUsage: rows, total = list_tags_with_usage(session, include_zero=False) - tag_names = {name for name, _, _ in rows} + tag_names = {name for name, _ in rows} assert "used" in tag_names assert "unused" not in tag_names @@ -262,7 +255,7 @@ class TestListTagsWithUsage: rows, total = list_tags_with_usage(session, prefix="alph") - tag_names = {name for name, _, _ in rows} + tag_names = {name for name, _ in rows} assert tag_names == {"alpha", "alphabet"} def test_order_by_name(self, session: Session): @@ -271,7 +264,7 @@ class TestListTagsWithUsage: rows, _ = list_tags_with_usage(session, order="name_asc") - names = [name for name, _, _ in rows] + names = [name for name, _ in rows] assert names == ["alpha", "middle", "zebra"] def test_owner_visibility(self, session: Session): @@ -287,13 +280,13 @@ class TestListTagsWithUsage: # Empty owner sees only shared rows, _ = list_tags_with_usage(session, owner_id="", include_zero=False) - tag_dict = {name: count for name, _, count in rows} + tag_dict = {name: count for name, count in rows} assert tag_dict.get("shared-tag", 0) == 1 assert tag_dict.get("owner-tag", 0) == 0 # User1 sees both rows, _ = list_tags_with_usage(session, owner_id="user1", include_zero=False) - tag_dict = {name: count for name, _, count in rows} + tag_dict = {name: count for name, count in rows} assert tag_dict.get("shared-tag", 0) == 1 assert tag_dict.get("owner-tag", 0) == 1 diff --git a/tests-unit/assets_test/services/test_tagging.py b/tests-unit/assets_test/services/test_tagging.py index ab69e5dc1..fa121db3e 100644 --- a/tests-unit/assets_test/services/test_tagging.py +++ b/tests-unit/assets_test/services/test_tagging.py @@ -141,7 +141,7 @@ class TestListTags: rows, total = list_tags() - tag_dict = {name: count for name, _, count in rows} + tag_dict = {name: count for name, count in rows} assert tag_dict["used"] == 1 assert tag_dict["unused"] == 0 assert total == 2 @@ -155,7 +155,7 @@ class TestListTags: rows, total = list_tags(include_zero=False) - tag_names = {name for name, _, _ in rows} + tag_names = {name for name, _ in rows} assert "used" in tag_names assert "unused" not in tag_names @@ -165,7 +165,7 @@ class TestListTags: rows, _ = list_tags(prefix="alph") - tag_names = {name for name, _, _ in rows} + tag_names = {name for name, _ in rows} assert tag_names == {"alpha", "alphabet"} def test_order_by_name(self, mock_create_session, session: Session): @@ -174,7 +174,7 @@ class TestListTags: rows, _ = list_tags(order="name_asc") - names = [name for name, _, _ in rows] + names = [name for name, _ in rows] assert names == ["alpha", "middle", "zebra"] def test_pagination(self, mock_create_session, session: Session): @@ -185,7 +185,7 @@ class TestListTags: assert total == 5 assert len(rows) == 2 - names = [name for name, _, _ in rows] + names = [name for name, _ in rows] assert names == ["b", "c"] def test_clamps_limit(self, mock_create_session, session: Session): diff --git a/tests-unit/assets_test/test_sync_references.py b/tests-unit/assets_test/test_sync_references.py index 94cc255bc..2e85076e0 100644 --- a/tests-unit/assets_test/test_sync_references.py +++ b/tests-unit/assets_test/test_sync_references.py @@ -95,7 +95,7 @@ def _make_asset( def _ensure_missing_tag(session: Session): """Ensure the 'missing' tag exists.""" if not session.get(Tag, "missing"): - session.add(Tag(name="missing", tag_type="system")) + session.add(Tag(name="missing")) session.flush() From 84e0692a3dd4748e197579da0e7fb0aa8510e363 Mon Sep 17 00:00:00 2001 From: Matt Miller Date: Tue, 9 Jun 2026 21:14:03 -0700 Subject: [PATCH 09/21] feat(assets): cursor-based pagination on GET /api/assets (#14014) MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit * spec(assets): add cursor pagination params to GET /api/assets Add 'after' query param and 'next_cursor' response field for keyset pagination. Matches the cloud Go implementation (BE-893) so frontend sees a unified contract across runtimes. Offset/limit remain as a deprecated fallback. * feat(assets): add cursor encode/decode helpers for keyset pagination Port of cloud common/pagination/cursor.go. Wire format is base64url of {"s", "v", "id"} JSON; times are Unix microseconds UTC to match PostgreSQL timestamp precision. Includes a byte-identity fixture pinned against the cloud Go wire format so cross-runtime FE pagination can't silently drift. * feat(assets): thread cursor through schemas, service, and query layer list_assets_page accepts an opaque 'after' cursor and returns next_cursor when more pages are available. The query applies a keyset WHERE clause and a secondary ORDER BY id for deterministic tiebreak. Cursor sort field is validated against the request sort, and a last_access_time sort (OSS-only) falls back to offset/limit. Offset is ignored whenever a cursor is supplied. * feat(assets): wire cursor pagination through GET /api/assets handler Adds integration tests for: full cursor walk, invalid-cursor 400, sort/cursor mismatch 400, cursor-wins-over-offset, absent next_cursor when no more results, and pagination stability across deletes. * fix(assets): address cursor-review verified findings - Mint next_cursor on every cursor-supported sort, not only when 'after' was supplied. A first request (no 'after') previously returned next_cursor=None, leaving cursor mode unreachable from a clean start. - Over-fetch limit+1 so an exactly-full terminal page doesn't mint a spurious cursor pointing at a phantom next page. - Map crafted out-of-range microsecond cursors (OverflowError / OSError in datetime construction) to 400 INVALID_CURSOR instead of leaking 500. - Bump MAX_CURSOR_VALUE_LENGTH 256 -> 512 to match the AssetReference name column max; without this, a long-named asset minted a cursor the same server then refused on the next request. Cross-runtime byte identity with cloud is unaffected because no cloud cursor ever carries a value > 256 (cloud schema doesn't permit it). - Return None from _encode_next_cursor when the boundary row carries a NULL sort value (e.g. an Asset without size_bytes backfilled), instead of silently encoding 0 and mis-positioning the keyset. - Fix schemas_in.py comment so it matches actual handler behavior (last_access_time + 'after' raises 400, does not fall back). - Add AssetsApiError schema + 400 response to GET /api/assets in openapi.yaml so generated clients know the INVALID_CURSOR envelope. - Extend integration coverage: first-page mint, exact-multiple terminal page, cursor walks for created_at/updated_at/size sorts, datetime overflow surfaces as 400 not 500. - Add unit coverage for datetime overflow and 512-char round-trip. * feat(assets): bind cursor to sort order + Go-compat JSON escaping Address three needs-judgment items from the cursor-review judge synthesis: 1. Cursor wire format now includes an "o" key carrying the sort direction ("asc" / "desc") it was minted under. A request that replays the cursor with a flipped `order` parameter is rejected with 400 INVALID_CURSOR instead of silently walking the wrong direction. Legacy cursors without "o" still decode (the binding is best-effort until cloud mirrors the field — follow-up filed separately). 2. JSON serialization now escapes `<`, `>`, `&`, U+2028, U+2029 to mirror Go's default `json.Marshal` behavior. Without this, an asset name containing those characters produced different bytes on Python vs cloud Go. The escaped form is what both runtimes emit. 3. Add direct query-layer tests for the keyset tiebreaker — the secondary ORDER BY id branch was previously unexercised. Two scenarios: all rows share a primary sort value, and mixed ties straddle page boundaries. Both assert no row is dropped or duplicated across the walk. Wire-format note: Python cursors now differ from current cloud cursors by exactly the "o" key. Cloud follow-up will bring the two back into byte alignment. * fix(assets): address bot review comments - Soften offset param prose: it's not deprecated, just not preferred for sequential walks. Random-access UIs (jump-to-page, item count displays) legitimately still want offset, so dropping the 'deprecated' framing rather than promoting it to a machine-readable deprecated:true flag. - Add explicit HTTP status assertions before every json() / next_cursor read in test_list_cursor.py so a failing request surfaces as an HTTP error instead of a confusing KeyError on a 4xx/5xx body. * feat(assets): require cursor o field, drop legacy permissive path Cursor pagination hasn't shipped on either runtime yet — this PR is still draft and cloud's mirror is just behind it — so there are no legacy no-o cursors in the wild. Make o mandatory from day one rather than landing permissive and tightening later. decode_cursor now rejects any payload without o (or with a non-string o) as malformed. CursorPayload.order becomes a required str. Tests that constructed CursorPayload directly now pass order="desc"; test_legacy_cursor_without_order_accepted flips to test_cursor_without_order_rejected. * chore(assets): drop cross-repo prose from cursor comments Strip prose references to sibling Go implementations and external ticket IDs from cursor.py, the cursor tests, the keyset integration tests, asset_management's sort-field comment, and the legacy prompt_id alias comment. Pure docstring/comment scrub — no behavior or wire-format changes. x-runtime: [cloud] field annotations in openapi.yaml are unchanged; those are the spec's structural cross-runtime convention, not internal references. * test(assets): include 'o' in microsecond-boundary cursor payload The boundary test was building a cursor without the required `o` key, so decode failed on the missing-order branch before reaching the µs-overflow path the test is asserting. Both paths return 400 INVALID_CURSOR so the assertion passed for the wrong reason. Add `o` to the payload and matching `order=` to the request so the decode reaches the intended branch. * fix(assets): address ultrareview findings on cursor pagination Six fact-checked findings from the multi-model review pass: - Encoder/decoder length asymmetry: encode_cursor now rejects empty id, oversized id (>128), oversized value (>512), and invalid order tokens symmetrically with decode_cursor. Prevents the same server from minting a cursor it then 400s on the next request (e.g. a filesystem-scanned asset name >512 chars). The bad-order path now raises InvalidCursorError (still subclasses ValueError) so route-layer handling stays uniform. - Raw U+2028/U+2029 in cursor.py source: ripgrep treated those lines as line-terminators, confirming the bytes were the actual separators. Any editor save / autoformat / git tooling that normalizes invisibles would silently break the encoder. Replaced with explicit 
 / 
 Python escape sequences. - set(seen) == set(names) hid ordering regressions: a cursor walk that dropped a row at a page boundary or returned duplicates could pass. Reworked the assertion to (1) reject duplicates, (2) require full coverage, and (3) assert strict positional order for size sort, the only field with a clock-independent ordering. - Flaky time.sleep(0.05) between inserts: Windows CI clock resolution is ~15ms, so back-to-back inserts under load could collide and exercise the tiebreaker instead of the documented path. Removed the sleep and let the strengthened assertion above carry coverage / no-duplicates, with size sort carrying strict order. - Cursor error envelope diverged from the rest of routes.py: cursor 400s emitted {error: {code, message}} while every other 400 in the file emits {error: {code, message, details}} via _build_error_response. Switched to _build_error_response and added the details field to the AssetsApiError schema in openapi.yaml. - "Byte-identity fixtures" only checked substring containment, defeating the test class's stated purpose of pinning the wire format. Switched to exact-bytes equality against an inline expected payload string per fixture, so any whitespace / key-order / escape drift fails loudly. Also dropped Go / json.Marshal references from docstrings — the byte format is the contract, not the runtime that mints it. * fix(assets): cap cursors by encoded wire size, not just char count Char-count guards on value/id can still let multibyte or escape-heavy inputs blow past MAX_ENCODED_CURSOR_LENGTH once UTF-8 + escape expansion + base64url runs. A 512-character name of 'é' (2 bytes UTF-8) or '<' (serializes to the 6-byte '<' escape) passes the char check, mints a ~1500-byte cursor, then 400s when handed back on the next request. Compute the final encoded form and reject it before returning if it exceeds the wire cap. Adds regression tests for both inflation paths. * refactor(assets): extract cursor JSON escaping helper; size wire cap above per-field caps Addresses review feedback on cursor.py: - Extract the inline escape chain into _apply_wire_compatible_json_escapes() with a comment pinning it to the wire format's escape set, so the parity intent is explicit rather than reading as an ad-hoc transform. - Raise MAX_ENCODED_CURSOR_LENGTH to 8192 (comfortably above the ~5.2KB worst-case the per-field caps can produce) and drop the mint-time length guard. Encoder/decoder symmetry now holds by construction: the encoder can't produce a cursor the decode path rejects, so there is no confusing user-visible 'cursor too long' failure at mint time. - Rewrite the two over-wire-cap tests to assert worst-case multibyte and escape-heavy values mint and round-trip, instead of being rejected. * refactor(assets): drop cross-runtime cursor escaping; cursors are opaque The custom JSON escaping of <, >, &, U+2028, and U+2029 existed only to keep the encoded cursor byte-identical with the Cloud implementation of the same payload format. Cursors are opaque tokens, so byte-level compatibility across implementations is not needed — plain json.dumps output is sufficient. Remove the escaping helper and the byte-identity test fixtures that pinned the wire format; keep round-trip coverage for the affected characters. --------- Co-authored-by: guill --- app/assets/api/routes.py | 40 +- app/assets/api/schemas_in.py | 5 + app/assets/api/schemas_out.py | 2 + .../database/queries/asset_reference.py | 35 +- app/assets/services/asset_management.py | 94 ++++- app/assets/services/cursor.py | 213 +++++++++++ app/assets/services/schemas.py | 1 + .../queries/test_asset_reference_keyset.py | 112 ++++++ .../assets_test/services/test_cursor.py | 278 ++++++++++++++ tests-unit/assets_test/test_list_cursor.py | 349 ++++++++++++++++++ 10 files changed, 1112 insertions(+), 17 deletions(-) create mode 100644 app/assets/services/cursor.py create mode 100644 tests-unit/assets_test/queries/test_asset_reference_keyset.py create mode 100644 tests-unit/assets_test/services/test_cursor.py create mode 100644 tests-unit/assets_test/test_list_cursor.py diff --git a/app/assets/api/routes.py b/app/assets/api/routes.py index 252ddfe8f..544a614f2 100644 --- a/app/assets/api/routes.py +++ b/app/assets/api/routes.py @@ -39,6 +39,7 @@ from app.assets.services import ( update_asset_metadata, upload_from_temp_path, ) +from app.assets.services.cursor import InvalidCursorError from app.assets.services.tagging import list_tag_histogram ROUTES = web.RouteTableDef() @@ -174,7 +175,7 @@ def _build_asset_response(result: schemas.AssetDetailResult | schemas.UploadResu user_metadata=result.ref.user_metadata or {}, metadata=result.ref.system_metadata, job_id=result.ref.job_id, - prompt_id=result.ref.job_id, # deprecated: mirrors job_id for cloud compat + prompt_id=result.ref.job_id, # deprecated alias of job_id, kept for compatibility created_at=result.ref.created_at, updated_at=result.ref.updated_at, last_access_time=result.ref.last_access_time, @@ -211,24 +212,37 @@ async def list_assets_route(request: web.Request) -> web.Response: order_candidate = (q.order or "desc").lower() order = order_candidate if order_candidate in {"asc", "desc"} else "desc" - result = list_assets_page( - owner_id=USER_MANAGER.get_request_user_id(request), - include_tags=q.include_tags, - exclude_tags=q.exclude_tags, - name_contains=q.name_contains, - metadata_filter=q.metadata_filter, - limit=q.limit, - offset=q.offset, - sort=sort, - order=order, - ) + try: + result = list_assets_page( + owner_id=USER_MANAGER.get_request_user_id(request), + include_tags=q.include_tags, + exclude_tags=q.exclude_tags, + name_contains=q.name_contains, + metadata_filter=q.metadata_filter, + limit=q.limit, + offset=q.offset, + sort=sort, + order=order, + after=q.after, + ) + except InvalidCursorError as e: + return _build_error_response(400, "INVALID_CURSOR", str(e)) summaries = [_build_asset_response(item) for item in result.items] + # has_more semantics differ by mode: + # - cursor mode: a non-empty next_cursor means there are more results. + # - offset mode: derived from total - (offset + page size). + if q.after is not None: + has_more = result.next_cursor is not None + else: + has_more = (q.offset + len(summaries)) < result.total + payload = schemas_out.AssetsList( assets=summaries, total=result.total, - has_more=(q.offset + len(summaries)) < result.total, + has_more=has_more, + next_cursor=result.next_cursor, ) return web.json_response(payload.model_dump(mode="json", exclude_none=True)) diff --git a/app/assets/api/schemas_in.py b/app/assets/api/schemas_in.py index 186a6ae1e..af666746d 100644 --- a/app/assets/api/schemas_in.py +++ b/app/assets/api/schemas_in.py @@ -59,6 +59,11 @@ class ListAssetsQuery(BaseModel): limit: conint(ge=1, le=500) = 20 offset: conint(ge=0) = 0 + # Opaque keyset cursor. When supplied, `offset` is ignored. Cursor pagination + # is supported for sort values `created_at`, `updated_at`, `name`, `size`. + # Supplying `after` together with `sort=last_access_time` returns + # 400 INVALID_CURSOR; that sort only supports offset/limit. + after: str | None = None sort: Literal["name", "created_at", "updated_at", "size", "last_access_time"] = ( "created_at" diff --git a/app/assets/api/schemas_out.py b/app/assets/api/schemas_out.py index 143848329..4e38e19d1 100644 --- a/app/assets/api/schemas_out.py +++ b/app/assets/api/schemas_out.py @@ -41,6 +41,8 @@ class AssetsList(BaseModel): assets: list[Asset] total: int has_more: bool + # Opaque cursor for the next page. Omitted when there are no more results. + next_cursor: str | None = None class TagUsage(BaseModel): diff --git a/app/assets/database/queries/asset_reference.py b/app/assets/database/queries/asset_reference.py index 8b90ae511..792411800 100644 --- a/app/assets/database/queries/asset_reference.py +++ b/app/assets/database/queries/asset_reference.py @@ -266,9 +266,18 @@ def list_references_page( metadata_filter: dict | None = None, sort: str | None = None, order: str | None = None, + after_cursor_value: object | None = None, + after_cursor_id: str | None = None, ) -> tuple[list[AssetReference], dict[str, list[str]], int]: """List references with pagination, filtering, and sorting. + When ``after_cursor_value``/``after_cursor_id`` are supplied the query uses + keyset pagination — ``offset`` is ignored and a WHERE clause selects rows + strictly after the given ``(sort_col, id)`` position in the active sort + direction. The cursor value must already be typed for the column + (datetime for time sorts, int for size, str for name); the caller decodes + the opaque cursor string and resolves to the typed value. + Returns (references, tag_map, total_count). """ base = ( @@ -297,9 +306,31 @@ def list_references_page( "size": Asset.size_bytes, } sort_col = sort_map.get(sort, AssetReference.created_at) - sort_exp = sort_col.desc() if order == "desc" else sort_col.asc() + descending = order == "desc" - base = base.order_by(sort_exp).limit(limit).offset(offset) + # Keyset WHERE: (sort_col, id) strictly less-than / greater-than the cursor. + # Equivalent to: sort_col v OR (sort_col = v AND id cursor_id). + if after_cursor_value is not None and after_cursor_id is not None: + if descending: + keyset = sa.or_( + sort_col < after_cursor_value, + sa.and_(sort_col == after_cursor_value, AssetReference.id < after_cursor_id), + ) + else: + keyset = sa.or_( + sort_col > after_cursor_value, + sa.and_(sort_col == after_cursor_value, AssetReference.id > after_cursor_id), + ) + base = base.where(keyset) + + # Secondary ORDER BY id (matching the primary direction) gives the keyset + # comparison a deterministic tiebreaker on duplicate sort_col values. + id_exp = AssetReference.id.desc() if descending else AssetReference.id.asc() + sort_exp = sort_col.desc() if descending else sort_col.asc() + + base = base.order_by(sort_exp, id_exp).limit(limit) + if after_cursor_id is None: + base = base.offset(offset) count_stmt = ( select(sa.func.count()) diff --git a/app/assets/services/asset_management.py b/app/assets/services/asset_management.py index 5aefd9956..1072c95fa 100644 --- a/app/assets/services/asset_management.py +++ b/app/assets/services/asset_management.py @@ -1,8 +1,19 @@ import contextlib import mimetypes import os +from datetime import timezone from typing import Sequence +from app.assets.services.cursor import ( + CursorPayload, + InvalidCursorError, + decode_cursor, + decode_cursor_int, + decode_cursor_time, + encode_cursor, + encode_cursor_from_time, +) + from app.assets.database.models import Asset from app.assets.database.queries import ( @@ -242,6 +253,11 @@ def get_asset_by_hash(asset_hash: str) -> AssetData | None: return extract_asset_data(asset) +# Sort fields that support cursor pagination. `last_access_time` is not +# in this list — it falls back to offset/limit. +_CURSOR_SORT_FIELDS = ("created_at", "updated_at", "name", "size") + + def list_assets_page( owner_id: str = "", include_tags: Sequence[str] | None = None, @@ -252,7 +268,39 @@ def list_assets_page( offset: int = 0, sort: str = "created_at", order: str = "desc", + after: str | None = None, ) -> ListAssetsResult: + """List assets with optional cursor pagination. + + When ``after`` is supplied it overrides ``offset``. The cursor's sort field + must match ``sort`` and be in the cursor-supported allowlist; mismatches + raise InvalidCursorError so the handler can map to 400 INVALID_CURSOR. + """ + cursor_value: object | None = None + cursor_id: str | None = None + # Mint next_cursor on every page where the sort is cursor-supported, not + # only when the request itself arrived with a cursor. Otherwise a first + # request (no `after`) returns next_cursor=None and the client can never + # enter cursor mode. + mint_cursor = sort in _CURSOR_SORT_FIELDS + + if after is not None: + if sort not in _CURSOR_SORT_FIELDS: + raise InvalidCursorError( + f"cursor pagination is not supported for sort={sort!r}" + ) + payload = decode_cursor(after, _CURSOR_SORT_FIELDS, expected_order=order) + if payload.sort_field != sort: + raise InvalidCursorError( + f"cursor sort field {payload.sort_field!r} does not match request sort {sort!r}" + ) + cursor_value, cursor_id = _resolve_cursor_value(payload), payload.id + + # Over-fetch by one row so we can distinguish "exactly `limit` rows total + # remaining" from "more rows past this page" without a second query. Drop + # the sentinel before returning. + fetch_limit = limit + 1 if mint_cursor else limit + with create_session() as session: refs, tag_map, total = list_references_page( session, @@ -261,12 +309,22 @@ def list_assets_page( exclude_tags=exclude_tags, name_contains=name_contains, metadata_filter=metadata_filter, - limit=limit, + limit=fetch_limit, offset=offset, sort=sort, order=order, + after_cursor_value=cursor_value, + after_cursor_id=cursor_id, ) + next_cursor: str | None = None + if mint_cursor and len(refs) > limit: + # There's at least one more row past this page — mint a cursor from + # the last row of the page (i.e. index `limit - 1`, since we + # over-fetched), and drop the sentinel. + next_cursor = _encode_next_cursor(refs[limit - 1], sort, order) + refs = refs[:limit] + items: list[AssetSummaryData] = [] for ref in refs: items.append( @@ -277,7 +335,39 @@ def list_assets_page( ) ) - return ListAssetsResult(items=items, total=total) + return ListAssetsResult(items=items, total=total, next_cursor=next_cursor) + + +def _resolve_cursor_value(payload: CursorPayload) -> object: + """Map a decoded cursor payload to a column-typed Python value.""" + if payload.sort_field in ("created_at", "updated_at"): + # DB stores naive UTC; strip tzinfo so the comparison binds against a + # `TIMESTAMP WITHOUT TIME ZONE` column without an offset shift. + return decode_cursor_time(payload).replace(tzinfo=None) + if payload.sort_field == "size": + return decode_cursor_int(payload) + return payload.value # name, str-typed + + +def _encode_next_cursor(ref, sort: str, order: str) -> str | None: + """Mint a cursor pointing at *ref* for the given sort dimension. + + Returns None when the boundary row carries a NULL sort value (e.g. an asset + record whose size_bytes hasn't been backfilled). Continuing pagination + across a NULL boundary is undefined under keyset ordering — better to + truncate cleanly here than to mint a cursor that mis-positions. + """ + if sort == "name": + return encode_cursor("name", ref.name, ref.id, order=order) + if sort == "size": + if ref.asset is None or ref.asset.size_bytes is None: + return None + return encode_cursor("size", str(ref.asset.size_bytes), ref.id, order=order) + # created_at / updated_at — DB datetimes are naive UTC; attach tz before encoding. + value = ref.created_at if sort == "created_at" else ref.updated_at + if value is None: + return None + return encode_cursor_from_time(sort, value.replace(tzinfo=timezone.utc), ref.id, order=order) def resolve_hash_to_path( diff --git a/app/assets/services/cursor.py b/app/assets/services/cursor.py new file mode 100644 index 000000000..6c7791528 --- /dev/null +++ b/app/assets/services/cursor.py @@ -0,0 +1,213 @@ +"""Opaque keyset-pagination cursor for /api/assets. + +Payload JSON uses short keys to keep the encoded length small: + + {"s": , "v": , "id": , "o": } + +The `o` key binds the cursor to the sort direction it was minted under, +so replaying a `desc` cursor against an `asc` request fails with +``INVALID_CURSOR`` rather than silently walking the wrong direction. +`o` is mandatory on every payload — a cursor without it is rejected as +malformed. + +Encoding is base64url with no padding. Cursors are opaque tokens: the +payload format is internal to this server, and clients must treat a +cursor as a black box handed back via `next_cursor`. No byte-level +compatibility with any other implementation is required. + +Time values are serialized as Unix microseconds (UTC) — microsecond +precision is sufficient to round-trip the timestamps stored by the +database without rounding rows in the same millisecond bucket. +""" +from __future__ import annotations + +import base64 +import json +from dataclasses import dataclass +from datetime import datetime, timezone +from typing import Iterable, Optional + + +class InvalidCursorError(ValueError): + """Raised on a malformed, oversized, or unsupported-sort-field cursor. + + Map to a 400 response with code ``INVALID_CURSOR`` at the handler. + """ + + +# Wire-format length caps. Cursors are user-controlled, so caps protect the +# decode path from oversized allocations and downstream SQL predicates from +# unbounded strings. +# +# MAX_CURSOR_VALUE_LENGTH is 512 to fit the `AssetReference.name` column max +# (`String(512)`) — otherwise a long-named asset would mint a cursor the same +# server then refuses on the next request. +# +# MAX_ENCODED_CURSOR_LENGTH is the decode-path guard, sized comfortably above +# the largest cursor the per-field caps can produce. Worst case is value + id +# at their caps with every character JSON-escaping to the six-byte `\uXXXX` +# form (control characters), which is ~5.2 KB once base64url-encoded. At 8192 +# the encoder can never mint a cursor that exceeds it, so a freshly minted +# cursor always decodes on the next request and there is no user-visible +# "cursor too long" failure. +MAX_ENCODED_CURSOR_LENGTH = 8192 +MAX_CURSOR_VALUE_LENGTH = 512 +MAX_CURSOR_ID_LENGTH = 128 + + +@dataclass(frozen=True) +class CursorPayload: + sort_field: str + value: str + id: str + order: str + + +_VALID_ORDERS = ("asc", "desc") + + +def encode_cursor(sort_field: str, value: str, id: str, order: str = "desc") -> str: + """Encode a cursor payload as a base64url (no-padding) string. + + `order` binds the cursor to the sort direction it was minted under so a + later request with a flipped `order` query parameter is rejected with + ``INVALID_CURSOR`` rather than silently walking the wrong direction. + """ + if order not in _VALID_ORDERS: + raise InvalidCursorError(f"order must be one of {_VALID_ORDERS}, got {order!r}") + # Symmetric input validation: the encoder must reject anything the + # decoder rejects, or the same server will mint cursors it then 400s on + # the next request. + if not id: + raise InvalidCursorError("id must be non-empty") + if len(id) > MAX_CURSOR_ID_LENGTH: + raise InvalidCursorError("id exceeds maximum length") + if len(value) > MAX_CURSOR_VALUE_LENGTH: + raise InvalidCursorError("value exceeds maximum length") + payload = {"s": sort_field, "v": value, "id": id, "o": order} + raw = json.dumps(payload, separators=(",", ":"), ensure_ascii=False) + # No mint-time length guard is needed: the per-field caps above bound the + # encoded length well below MAX_ENCODED_CURSOR_LENGTH (see its definition), + # so the encoder can never produce a cursor the decode path would reject. + return base64.urlsafe_b64encode(raw.encode("utf-8")).rstrip(b"=").decode("ascii") + + +def encode_cursor_from_time(sort_field: str, t: datetime, id: str, order: str = "desc") -> str: + """Encode a time-typed cursor at Unix microsecond precision. + + Accepts an aware datetime (any timezone) and normalizes to UTC. Naive + datetimes are rejected so callers can't accidentally encode the local + wall-clock value of a UTC-stored timestamp. + """ + if t.tzinfo is None: + raise ValueError("encode_cursor_from_time requires an aware datetime") + micros = _datetime_to_unix_micros(t.astimezone(timezone.utc)) + return encode_cursor(sort_field, str(micros), id, order=order) + + +def decode_cursor( + cursor: str, + allowed_sort_fields: Iterable[str], + expected_order: str | None = None, +) -> CursorPayload: + """Parse an opaque cursor. + + ``allowed_sort_fields`` is the endpoint's accepted sort-field list — a + cursor carrying a field outside this set is rejected so a cursor minted + for one column can't be replayed against another (e.g. a ``created_at`` + timestamp string compared against a ``name`` column). + + ``expected_order`` (``"asc"``/``"desc"``), when supplied, must match the + payload's ``o`` field. ``o`` is required on every payload; a cursor + missing it is rejected as malformed. + + Passing no allowed fields rejects every cursor. + """ + if len(cursor) > MAX_ENCODED_CURSOR_LENGTH: + raise InvalidCursorError("cursor exceeds maximum length") + + try: + # urlsafe_b64decode requires correct padding; we strip on encode, so + # restore the trailing '=' pad here. + padding = "=" * (-len(cursor) % 4) + raw = base64.urlsafe_b64decode(cursor + padding) + except (ValueError, base64.binascii.Error) as e: + raise InvalidCursorError(f"encoding: {e}") from e + + try: + decoded = json.loads(raw) + except (json.JSONDecodeError, UnicodeDecodeError) as e: + raise InvalidCursorError(f"payload: {e}") from e + + if not isinstance(decoded, dict): + raise InvalidCursorError("payload: expected object") + + sort_field = decoded.get("s") + value = decoded.get("v") + id = decoded.get("id") + order = decoded.get("o") + + if not isinstance(sort_field, str) or not isinstance(value, str) or not isinstance(id, str): + raise InvalidCursorError("payload: missing or non-string s/v/id") + + if id == "": + raise InvalidCursorError("missing id") + if len(id) > MAX_CURSOR_ID_LENGTH: + raise InvalidCursorError("id exceeds maximum length") + if len(value) > MAX_CURSOR_VALUE_LENGTH: + raise InvalidCursorError("value exceeds maximum length") + + if sort_field not in allowed_sort_fields: + raise InvalidCursorError(f"unsupported sort field {sort_field!r}") + + if not isinstance(order, str): + raise InvalidCursorError("missing or non-string o") + if order not in _VALID_ORDERS: + raise InvalidCursorError(f"unsupported order {order!r}") + if expected_order is not None and order != expected_order: + raise InvalidCursorError( + f"cursor order {order!r} does not match request order {expected_order!r}" + ) + + return CursorPayload(sort_field=sort_field, value=value, id=id, order=order) + + +def decode_cursor_time(payload: Optional[CursorPayload]) -> datetime: + """Parse a time-typed cursor value as Unix microseconds, returning UTC.""" + if payload is None: + raise InvalidCursorError("nil cursor payload") + try: + micros = int(payload.value) + except ValueError as e: + raise InvalidCursorError(f"value is not a valid timestamp: {e}") from e + try: + return _unix_micros_to_datetime(micros) + except (OverflowError, OSError, ValueError) as e: + # Crafted out-of-range microseconds (e.g. > datetime.MAX_YEAR) blow up + # in fromtimestamp / datetime construction. Map to 400, not 500. + raise InvalidCursorError(f"value is out of representable range: {e}") from e + + +def decode_cursor_int(payload: Optional[CursorPayload]) -> int: + """Parse a cursor value as a base-10 integer.""" + if payload is None: + raise InvalidCursorError("nil cursor payload") + try: + return int(payload.value) + except ValueError as e: + raise InvalidCursorError(f"value is not a valid integer: {e}") from e + + +_EPOCH = datetime(1970, 1, 1, tzinfo=timezone.utc) + + +def _datetime_to_unix_micros(t: datetime) -> int: + """Convert an aware UTC datetime to Unix microseconds (integer math).""" + delta = t - _EPOCH + return (delta.days * 86_400 + delta.seconds) * 1_000_000 + delta.microseconds + + +def _unix_micros_to_datetime(micros: int) -> datetime: + """Convert Unix microseconds to a UTC datetime, preserving precision.""" + seconds, micro_remainder = divmod(micros, 1_000_000) + return datetime.fromtimestamp(seconds, tz=timezone.utc).replace(microsecond=micro_remainder) diff --git a/app/assets/services/schemas.py b/app/assets/services/schemas.py index 2a52d76ca..4d2af8a02 100644 --- a/app/assets/services/schemas.py +++ b/app/assets/services/schemas.py @@ -70,6 +70,7 @@ class AssetSummaryData: class ListAssetsResult: items: list[AssetSummaryData] total: int + next_cursor: str | None = None @dataclass(frozen=True) diff --git a/tests-unit/assets_test/queries/test_asset_reference_keyset.py b/tests-unit/assets_test/queries/test_asset_reference_keyset.py new file mode 100644 index 000000000..d143d60f9 --- /dev/null +++ b/tests-unit/assets_test/queries/test_asset_reference_keyset.py @@ -0,0 +1,112 @@ +"""Keyset-pagination tiebreaker tests for list_references_page. + +When multiple rows share the same primary sort value (e.g. four assets +created in the same microsecond), the secondary `ORDER BY id` is what keeps +keyset pagination from losing or repeating rows. This file exercises that +branch directly against an in-memory SQLite session — engineering identical +timestamps via HTTP is unreliable enough that we work at the query layer. +""" +import uuid +from datetime import datetime + +import pytest +from sqlalchemy.orm import Session + +from app.assets.database.models import Asset, AssetReference +from app.assets.database.queries.asset_reference import list_references_page + + +def _make_ref(session: Session, created_at: datetime, name: str, owner: str = "") -> AssetReference: + asset = Asset(hash=f"blake3:{uuid.uuid4().hex}", size_bytes=1024) + session.add(asset) + session.flush() + ref = AssetReference( + id=str(uuid.uuid4()), + asset_id=asset.id, + owner_id=owner, + name=name, + file_path=f"/tmp/{name}", + created_at=created_at, + updated_at=created_at, + last_access_time=created_at, + is_missing=False, + ) + session.add(ref) + return ref + + +@pytest.mark.parametrize("order", ["desc", "asc"]) +def test_tiebreaker_walks_duplicate_sort_values(session: Session, order: str): + """Four rows with the SAME created_at must paginate cleanly under cursor + mode — no row dropped, no row repeated, despite the primary sort column + being non-discriminating. + """ + shared_ts = datetime(2024, 5, 20, 12, 0, 0) # naive UTC, like the DB stores + refs = [_make_ref(session, shared_ts, f"tie_{i}.png") for i in range(4)] + session.commit() + + expected_ids = sorted([r.id for r in refs], reverse=(order == "desc")) + + # Walk the cursor by hand: page size 2, take 3 pages (2 + 2 + 0). + seen: list[str] = [] + after_value = None + after_id = None + for _ in range(4): # generous loop bound; ought to be 2 iterations + page, _tag_map, _total = list_references_page( + session, + limit=2, + sort="created_at", + order=order, + after_cursor_value=after_value, + after_cursor_id=after_id, + ) + if not page: + break + seen.extend(p.id for p in page) + # Use the last row's (created_at, id) as the next cursor input. + last = page[-1] + after_value, after_id = last.created_at, last.id + if len(page) < 2: + break + + assert seen == expected_ids, ( + f"keyset tiebreaker failed for order={order}: expected {expected_ids}, got {seen}" + ) + + +def test_tiebreaker_no_duplicates_under_mixed_collisions(session: Session): + """Some rows share a timestamp, some don't. The cursor must still walk + every row exactly once regardless of where ties sit relative to a + page boundary.""" + t1 = datetime(2024, 5, 20, 12, 0, 0) + t2 = datetime(2024, 5, 20, 12, 0, 1) + layout = [t1, t1, t1, t2, t2] # three rows at t1, two at t2 + refs = [_make_ref(session, ts, f"mix_{i}.png") for i, ts in enumerate(layout)] + session.commit() + + all_ids = {r.id for r in refs} + seen_set: set[str] = set() + seen_list: list[str] = [] + after_value = None + after_id = None + for _ in range(6): + page, _, _ = list_references_page( + session, + limit=2, + sort="created_at", + order="desc", + after_cursor_value=after_value, + after_cursor_id=after_id, + ) + if not page: + break + for p in page: + assert p.id not in seen_set, f"duplicate row {p.id} appeared in cursor walk" + seen_set.add(p.id) + seen_list.append(p.id) + last = page[-1] + after_value, after_id = last.created_at, last.id + if len(page) < 2: + break + + assert seen_set == all_ids, f"missing rows: expected {all_ids}, got {seen_set}" diff --git a/tests-unit/assets_test/services/test_cursor.py b/tests-unit/assets_test/services/test_cursor.py new file mode 100644 index 000000000..47970e168 --- /dev/null +++ b/tests-unit/assets_test/services/test_cursor.py @@ -0,0 +1,278 @@ +"""Tests for app.assets.services.cursor. + +Cursors are opaque tokens internal to this server — these tests cover +round-tripping, validation, and length caps, not any particular wire +byte layout. +""" +from __future__ import annotations + +import base64 +from datetime import datetime, timedelta, timezone + +import pytest + +from app.assets.services.cursor import ( + MAX_CURSOR_ID_LENGTH, + MAX_CURSOR_VALUE_LENGTH, + MAX_ENCODED_CURSOR_LENGTH, + CursorPayload, + InvalidCursorError, + decode_cursor, + decode_cursor_int, + decode_cursor_time, + encode_cursor, + encode_cursor_from_time, +) + + +ALLOWED = ("created_at", "updated_at", "name", "size") + + +class TestRoundTrip: + @pytest.mark.parametrize( + "sort_field, value, id", + [ + ("created_at", "1716200000000000", "a1b2c3d4-e5f6-7a89-b0c1-d2e3f4a5b6c7"), + ("size", "1024", "asset-123"), + ("name", "my-asset.png", "asset-abc"), + ("name", "résumé.txt", "asset-uni"), + ("name", "foo<&>bar.png", "asset-html"), + ("name", 'quo"te\\back\nnewline.png', "asset-esc"), + ], + ) + def test_encode_decode(self, sort_field, value, id): + encoded = encode_cursor(sort_field, value, id) + assert encoded != "" + payload = decode_cursor(encoded, ALLOWED) + assert payload.sort_field == sort_field + assert payload.value == value + assert payload.id == id + + +class TestTimeCursor: + def test_microsecond_precision_preserved(self): + # Pick a time with non-zero microseconds — encoding at ms would lose the µs. + ts = datetime(2024, 5, 20, 12, 53, 20, 123456, tzinfo=timezone.utc) + encoded = encode_cursor_from_time("created_at", ts, "id-1") + payload = decode_cursor(encoded, ALLOWED) + # Value must be a microsecond integer string, not a millisecond one. + assert payload.value == "1716209600123456" + decoded = decode_cursor_time(payload) + assert decoded == ts + + def test_decode_returns_utc(self): + payload = CursorPayload(sort_field="created_at", value="1716200000123456", id="id-1", order="desc") + decoded = decode_cursor_time(payload) + assert decoded.tzinfo == timezone.utc + + def test_naive_datetime_rejected_on_encode(self): + naive = datetime(2024, 5, 20, 12, 0, 0) + with pytest.raises(ValueError): + encode_cursor_from_time("created_at", naive, "id-1") + + def test_non_integer_value_rejected_on_decode(self): + with pytest.raises(InvalidCursorError): + decode_cursor_time(CursorPayload("created_at", "not-a-number", "id-1", "desc")) + + def test_none_payload_rejected(self): + with pytest.raises(InvalidCursorError): + decode_cursor_time(None) + + def test_non_utc_aware_normalized(self): + # Same instant, different timezone — must encode to the same micros. + utc_ts = datetime(2024, 5, 20, 12, 0, 0, tzinfo=timezone.utc) + offset_ts = utc_ts.astimezone(timezone(timedelta(hours=-5))) + assert encode_cursor_from_time("created_at", utc_ts, "x") == encode_cursor_from_time( + "created_at", offset_ts, "x" + ) + + +class TestIntCursor: + def test_decode_int(self): + assert decode_cursor_int(CursorPayload("size", "1024", "id-1", "desc")) == 1024 + + def test_decode_int_rejects_non_int(self): + with pytest.raises(InvalidCursorError): + decode_cursor_int(CursorPayload("size", "abc", "id-1", "desc")) + + def test_decode_int_rejects_none(self): + with pytest.raises(InvalidCursorError): + decode_cursor_int(None) + + +class TestInvalidInputs: + def test_oversized_cursor(self): + oversized = "a" * (MAX_ENCODED_CURSOR_LENGTH + 1) + with pytest.raises(InvalidCursorError, match="maximum length"): + decode_cursor(oversized, ALLOWED) + + def test_not_base64(self): + with pytest.raises(InvalidCursorError): + decode_cursor("not base64!!!", ALLOWED) + + def test_not_json(self): + encoded = base64.urlsafe_b64encode(b"definitely not json").rstrip(b"=").decode("ascii") + with pytest.raises(InvalidCursorError): + decode_cursor(encoded, ALLOWED) + + def test_empty_id(self): + # Encoder rejects empty id symmetrically with the decoder, so build the + # payload manually to exercise the decoder's missing-id branch. + raw = b'{"s":"created_at","v":"1","id":"","o":"desc"}' + encoded = base64.urlsafe_b64encode(raw).rstrip(b"=").decode("ascii") + with pytest.raises(InvalidCursorError, match="missing id"): + decode_cursor(encoded, ALLOWED) + + def test_oversized_id(self): + # Encoder enforces the cap symmetrically; hand-build to exercise decode. + big_id = "a" * (MAX_CURSOR_ID_LENGTH + 1) + raw = ('{"s":"created_at","v":"1","id":"' + big_id + '","o":"desc"}').encode("ascii") + encoded = base64.urlsafe_b64encode(raw).rstrip(b"=").decode("ascii") + with pytest.raises(InvalidCursorError, match="id exceeds maximum length"): + decode_cursor(encoded, ALLOWED) + + def test_oversized_value(self): + # Encoder enforces the cap symmetrically; hand-build to exercise decode. + big_v = "v" * (MAX_CURSOR_VALUE_LENGTH + 1) + raw = ('{"s":"created_at","v":"' + big_v + '","id":"id-1","o":"desc"}').encode("ascii") + encoded = base64.urlsafe_b64encode(raw).rstrip(b"=").decode("ascii") + with pytest.raises(InvalidCursorError, match="value exceeds maximum length"): + decode_cursor(encoded, ALLOWED) + + def test_unsupported_sort_field(self): + encoded = encode_cursor("execution_time", "1", "id-1") + with pytest.raises(InvalidCursorError, match="unsupported sort field"): + decode_cursor(encoded, ALLOWED) + + def test_no_allowed_fields_rejects_everything(self): + encoded = encode_cursor("created_at", "1", "id-1") + with pytest.raises(InvalidCursorError): + decode_cursor(encoded, ()) + + def test_non_dict_payload_rejected(self): + encoded = base64.urlsafe_b64encode(b'["array","not","dict"]').rstrip(b"=").decode("ascii") + with pytest.raises(InvalidCursorError, match="expected object"): + decode_cursor(encoded, ALLOWED) + + +class TestEncodeAtCapsFits: + def test_max_field_lengths_fit_wire_cap(self): + # Worst-case payload: value and id at their per-field caps, with a long + # sort field name. The encoded cursor must fit within MAX_ENCODED_CURSOR_LENGTH + # so the wire cap cannot reject a cursor the encoder mints at the per-field caps. + value = "v" * MAX_CURSOR_VALUE_LENGTH + id = "i" * MAX_CURSOR_ID_LENGTH + sort_field = "very_long_sort_field_name" + + encoded = encode_cursor(sort_field, value, id) + assert len(encoded) <= MAX_ENCODED_CURSOR_LENGTH + payload = decode_cursor(encoded, (sort_field,)) + assert payload.value == value + assert payload.id == id + + +class TestDatetimeOverflow: + """Crafted cursors with extreme micros must map to InvalidCursorError, + not OverflowError/OSError leaking as 500. + """ + + @pytest.mark.parametrize( + "micros_str", + [ + "999999999999999999999", # 10^21 µs — past datetime.MAX_YEAR by ~14 orders + "-999999999999999999999", # symmetric negative — pre-epoch overflow + ], + ) + def test_out_of_range_micros_rejected(self, micros_str): + encoded = encode_cursor("created_at", micros_str, "asset-x") + payload = decode_cursor(encoded, ALLOWED) + with pytest.raises(InvalidCursorError): + decode_cursor_time(payload) + + +class TestEncoderDecoderSymmetry: + """The encoder must never mint a cursor the decoder would reject, or the + same server would 400 on a cursor it just handed out. Per-field caps keep + the encoded length below the wire cap, so a freshly minted cursor always + round-trips. + """ + + def test_long_name_within_cap_round_trips(self): + """Assets allow names up to 512 chars (`String(512)`); the cursor + encoder must round-trip a value at that cap so a freshly minted + cursor never fails decode on the next request.""" + long_name = "n" * MAX_CURSOR_VALUE_LENGTH + encoded = encode_cursor("name", long_name, "asset-x") + payload = decode_cursor(encoded, ALLOWED) + assert payload.value == long_name + + def test_encoder_rejects_empty_id(self): + with pytest.raises(InvalidCursorError, match="id must be non-empty"): + encode_cursor("created_at", "1", "") + + def test_encoder_rejects_oversized_id(self): + with pytest.raises(InvalidCursorError, match="id exceeds maximum length"): + encode_cursor("created_at", "1", "a" * (MAX_CURSOR_ID_LENGTH + 1)) + + def test_encoder_rejects_oversized_value(self): + with pytest.raises(InvalidCursorError, match="value exceeds maximum length"): + encode_cursor("name", "v" * (MAX_CURSOR_VALUE_LENGTH + 1), "id-1") + + def test_multibyte_value_at_cap_round_trips(self): + """A value at the char-count cap made of multibyte characters + (e.g. 'é' = 2 UTF-8 bytes) stays under the wire cap, so it mints and + round-trips — the per-field caps, not a mint-time length check, are + what bound cursor size.""" + value = "é" * MAX_CURSOR_VALUE_LENGTH + encoded = encode_cursor("name", value, "asset-multibyte") + assert len(encoded) <= MAX_ENCODED_CURSOR_LENGTH + payload = decode_cursor(encoded, ALLOWED) + assert payload.value == value + + def test_escape_heavy_value_at_cap_round_trips(self): + """JSON escape expansion is the worst case: each control character + serializes to the six-byte `\\uXXXX` form. A value of 512 of them is + the largest a cursor can get, and it still fits the wire cap, mints, + and round-trips.""" + value = "\x01" * MAX_CURSOR_VALUE_LENGTH + encoded = encode_cursor("name", value, "asset-escape") + assert len(encoded) <= MAX_ENCODED_CURSOR_LENGTH + payload = decode_cursor(encoded, ALLOWED) + assert payload.value == value + + +class TestOrderBinding: + def test_order_baked_into_payload(self): + encoded = encode_cursor("created_at", "1", "id-1", order="asc") + payload = decode_cursor(encoded, ALLOWED) + assert payload.order == "asc" + + def test_mismatched_order_rejected(self): + encoded = encode_cursor("created_at", "1", "id-1", order="desc") + with pytest.raises(InvalidCursorError, match="does not match request order"): + decode_cursor(encoded, ALLOWED, expected_order="asc") + + def test_matching_order_accepted(self): + encoded = encode_cursor("created_at", "1", "id-1", order="desc") + payload = decode_cursor(encoded, ALLOWED, expected_order="desc") + assert payload.order == "desc" + + def test_invalid_order_token_rejected_on_encode(self): + with pytest.raises(ValueError): + encode_cursor("created_at", "1", "id-1", order="sideways") + + def test_invalid_order_token_rejected_on_decode(self): + # Hand-craft a payload with an illegal `o` value. + raw = b'{"s":"name","v":"x","id":"id-1","o":"sideways"}' + encoded = base64.urlsafe_b64encode(raw).rstrip(b"=").decode("ascii") + with pytest.raises(InvalidCursorError, match="unsupported order"): + decode_cursor(encoded, ALLOWED) + + def test_cursor_without_order_rejected(self): + """`o` is mandatory. A cursor minted without it is rejected as + malformed rather than silently walking the keyset in whatever + direction the request happens to ask for.""" + raw = b'{"s":"name","v":"x","id":"id-1"}' + encoded = base64.urlsafe_b64encode(raw).rstrip(b"=").decode("ascii") + with pytest.raises(InvalidCursorError, match="missing or non-string o"): + decode_cursor(encoded, ALLOWED, expected_order="desc") diff --git a/tests-unit/assets_test/test_list_cursor.py b/tests-unit/assets_test/test_list_cursor.py new file mode 100644 index 000000000..a37019fd6 --- /dev/null +++ b/tests-unit/assets_test/test_list_cursor.py @@ -0,0 +1,349 @@ +"""Integration tests for cursor-based pagination on GET /api/assets. + +These tests exercise the handler/service/query path end-to-end; +cursor-encoding-level tests live in +tests-unit/assets_test/services/test_cursor.py. +""" +import pytest +import requests + + +def _seed(asset_factory, make_asset_bytes, count: int, tag: str) -> list[str]: + names = [f"cursor_{i:02d}.safetensors" for i in range(count)] + for n in names: + asset_factory( + n, + ["models", "checkpoints", "unit-tests", tag], + {}, + make_asset_bytes(n, size=2048), + ) + return sorted(names) + + +def test_cursor_pages_all_items_in_order(http: requests.Session, api_base: str, asset_factory, make_asset_bytes): + names = _seed(asset_factory, make_asset_bytes, count=5, tag="cursor-walk") + + params = { + "include_tags": "unit-tests,cursor-walk", + "sort": "name", + "order": "asc", + "limit": "2", + } + + seen: list[str] = [] + after: str | None = None + pages = 0 + while True: + page_params = dict(params) + if after is not None: + page_params["after"] = after + r = http.get(api_base + "/api/assets", params=page_params, timeout=120) + assert r.status_code == 200, r.text + body = r.json() + seen.extend(a["name"] for a in body["assets"]) + pages += 1 + after = body.get("next_cursor") + if after is None: + break + assert body["has_more"] is True + assert pages < 10, "guard against runaway cursor loop" + + assert seen == names, f"expected {names}, got {seen}" + # Last page should have has_more False + assert body["has_more"] is False + assert "next_cursor" not in body + + +def test_cursor_invalid_returns_400(http: requests.Session, api_base: str): + r = http.get( + api_base + "/api/assets", + params={"after": "not-a-real-cursor", "sort": "created_at"}, + timeout=120, + ) + assert r.status_code == 400, r.text + body = r.json() + assert body["error"]["code"] == "INVALID_CURSOR" + + +def test_cursor_sort_mismatch_returns_400(http: requests.Session, api_base: str, asset_factory, make_asset_bytes): + _seed(asset_factory, make_asset_bytes, count=2, tag="cursor-mismatch") + + # Take a real cursor minted for sort=name. + r = http.get( + api_base + "/api/assets", + params={ + "include_tags": "unit-tests,cursor-mismatch", + "sort": "name", + "order": "asc", + "limit": "1", + }, + timeout=120, + ) + assert r.status_code == 200 + cursor = r.json()["next_cursor"] + assert cursor is not None + + # Replay against sort=created_at — should fail with INVALID_CURSOR. + r2 = http.get( + api_base + "/api/assets", + params={"after": cursor, "sort": "created_at"}, + timeout=120, + ) + assert r2.status_code == 400, r2.text + assert r2.json()["error"]["code"] == "INVALID_CURSOR" + + +def test_cursor_wins_over_offset(http: requests.Session, api_base: str, asset_factory, make_asset_bytes): + names = _seed(asset_factory, make_asset_bytes, count=4, tag="cursor-vs-offset") + + # Take a cursor that points past the first item. + r = http.get( + api_base + "/api/assets", + params={ + "include_tags": "unit-tests,cursor-vs-offset", + "sort": "name", + "order": "asc", + "limit": "1", + }, + timeout=120, + ) + assert r.status_code == 200, r.text + cursor = r.json()["next_cursor"] + assert cursor is not None + + # Pass both 'after' and a large offset. Cursor must win; offset is ignored. + r2 = http.get( + api_base + "/api/assets", + params={ + "include_tags": "unit-tests,cursor-vs-offset", + "sort": "name", + "order": "asc", + "limit": "1", + "after": cursor, + "offset": "999", + }, + timeout=120, + ) + assert r2.status_code == 200 + body = r2.json() + # Should land on the second name in sorted order — not skip ahead by 999. + assert [a["name"] for a in body["assets"]] == [names[1]] + + +def test_next_cursor_absent_when_no_more_results(http: requests.Session, api_base: str, asset_factory, make_asset_bytes): + _seed(asset_factory, make_asset_bytes, count=2, tag="cursor-exhaust") + + r = http.get( + api_base + "/api/assets", + params={ + "include_tags": "unit-tests,cursor-exhaust", + "sort": "name", + "order": "asc", + "limit": "50", + }, + timeout=120, + ) + assert r.status_code == 200, r.text + body = r.json() + assert body["has_more"] is False + assert "next_cursor" not in body + + +def test_cursor_pagination_first_page_mints_cursor(http: requests.Session, api_base: str, asset_factory, make_asset_bytes): + """First-page request (no `after`) must still return `next_cursor` when + more rows exist, or pagination is unreachable from a cold start. + """ + _seed(asset_factory, make_asset_bytes, count=3, tag="cursor-first-page") + r = http.get( + api_base + "/api/assets", + params={"include_tags": "unit-tests,cursor-first-page", "sort": "name", "order": "asc", "limit": "2"}, + timeout=120, + ) + assert r.status_code == 200, r.text + body = r.json() + assert body["has_more"] is True + assert body.get("next_cursor"), "first page must mint a cursor when more rows exist" + + +def test_cursor_no_spurious_cursor_when_page_size_equals_remainder(http: requests.Session, api_base: str, asset_factory, make_asset_bytes): + """When `total` is an exact multiple of `limit`, the final page must + NOT carry a next_cursor — there is nothing past it. + """ + _seed(asset_factory, make_asset_bytes, count=4, tag="cursor-exact-multiple") + # Page 1 + r = http.get( + api_base + "/api/assets", + params={"include_tags": "unit-tests,cursor-exact-multiple", "sort": "name", "order": "asc", "limit": "2"}, + timeout=120, + ) + assert r.status_code == 200, r.text + cursor = r.json()["next_cursor"] + assert cursor is not None + # Page 2 — should exhaust the set with no cursor for a phantom page 3 + r2 = http.get( + api_base + "/api/assets", + params={"include_tags": "unit-tests,cursor-exact-multiple", "sort": "name", "order": "asc", "limit": "2", "after": cursor}, + timeout=120, + ) + assert r2.status_code == 200, r2.text + body = r2.json() + assert len(body["assets"]) == 2 + assert body["has_more"] is False + assert "next_cursor" not in body + + +@pytest.mark.parametrize("sort_field", ["created_at", "updated_at", "size"]) +def test_cursor_walks_for_non_name_sorts(sort_field, http: requests.Session, api_base: str, asset_factory, make_asset_bytes): + """Cursor pagination must work for every sort field the contract claims. + + Without this, the `created_at` / `updated_at` (time-encoded micros) and + `size` (int-encoded) cursor paths go entirely unexercised end-to-end. + """ + # Sizes increase strictly by index, so `size desc` has a deterministic + # expected order. Time-based sorts (created_at / updated_at) can tie when + # rows are inserted faster than the DB's timestamp resolution; for those + # we check coverage and no-duplicates and let the keyset tiebreaker do + # the rest, instead of sleeping between inserts and asserting an order + # that depends on clock granularity. + names = [] + for i in range(4): + n = f"cursor_{sort_field}_{i:02d}.safetensors" + asset_factory(n, ["models", "checkpoints", "unit-tests", f"cursor-{sort_field}"], {}, make_asset_bytes(n, size=2048 + i)) + names.append(n) + + params = { + "include_tags": f"unit-tests,cursor-{sort_field}", + "sort": sort_field, + "order": "desc", + "limit": "2", + } + seen: list[str] = [] + after: str | None = None + pages = 0 + while True: + page_params = dict(params) + if after is not None: + page_params["after"] = after + r = http.get(api_base + "/api/assets", params=page_params, timeout=120) + assert r.status_code == 200, r.text + body = r.json() + seen.extend(a["name"] for a in body["assets"]) + after = body.get("next_cursor") + pages += 1 + if after is None: + break + assert pages < 10, "guard against runaway cursor loop" + + # No duplicates: a faulty keyset boundary that returns the same row across + # two pages must fail this check. + assert len(seen) == len(set(seen)), ( + f"cursor walk repeated rows for sort={sort_field}: {seen}" + ) + # Full coverage: every seeded asset reached exactly once. + assert set(seen) == set(names), ( + f"missing items for sort={sort_field}: expected {set(names)}, got {set(seen)}" + ) + # Strict order check for the only field with a clock-independent ordering. + if sort_field == "size": + assert seen == list(reversed(names)), ( + f"size cursor walked out of order: got {seen}, expected {list(reversed(names))}" + ) + + +def test_cursor_order_mismatch_returns_400(http: requests.Session, api_base: str, asset_factory, make_asset_bytes): + """A cursor minted under desc order replayed against asc must 400, not + silently walk the wrong direction.""" + _seed(asset_factory, make_asset_bytes, count=3, tag="cursor-order-flip") + + r = http.get( + api_base + "/api/assets", + params={ + "include_tags": "unit-tests,cursor-order-flip", + "sort": "name", + "order": "desc", + "limit": "1", + }, + timeout=120, + ) + assert r.status_code == 200, r.text + cursor = r.json()["next_cursor"] + assert cursor is not None + + # Replay with order flipped to asc — server must reject the cursor. + r2 = http.get( + api_base + "/api/assets", + params={ + "include_tags": "unit-tests,cursor-order-flip", + "sort": "name", + "order": "asc", + "limit": "1", + "after": cursor, + }, + timeout=120, + ) + assert r2.status_code == 400, r2.text + assert r2.json()["error"]["code"] == "INVALID_CURSOR" + + +def test_cursor_invalid_cursor_at_microsecond_boundary(http: requests.Session, api_base: str): + """A cursor carrying an out-of-range microsecond timestamp must map to + 400 INVALID_CURSOR, not 500.""" + import base64 + import json + # 10^18 microseconds ≈ year 33658, well past datetime.MAX_YEAR. + # `o` and `order=` must be set; otherwise decode fails earlier on the + # missing-order branch and the µs-overflow path is never exercised. + payload = {"s": "created_at", "o": "desc", "v": "999999999999999999999", "id": "asset-x"} + raw = json.dumps(payload, separators=(",", ":")).encode("utf-8") + cursor = base64.urlsafe_b64encode(raw).rstrip(b"=").decode("ascii") + r = http.get( + api_base + "/api/assets", + params={"after": cursor, "sort": "created_at", "order": "desc"}, + timeout=120, + ) + assert r.status_code == 400, r.text + assert r.json()["error"]["code"] == "INVALID_CURSOR" + + +def test_cursor_pagination_stable_after_delete(http: requests.Session, api_base: str, asset_factory, make_asset_bytes): + names = _seed(asset_factory, make_asset_bytes, count=4, tag="cursor-delete") + + # Page 1. + r = http.get( + api_base + "/api/assets", + params={ + "include_tags": "unit-tests,cursor-delete", + "sort": "name", + "order": "asc", + "limit": "2", + }, + timeout=120, + ) + assert r.status_code == 200 + body = r.json() + page1_names = [a["name"] for a in body["assets"]] + cursor = body["next_cursor"] + assert cursor is not None + assert page1_names == names[:2] + + # Delete an item from page 1 (already returned) — cursor should still + # locate the next page from where it was minted, not re-index. + target_id = body["assets"][0]["id"] + d = http.delete(api_base + f"/api/assets/{target_id}", timeout=120) + assert d.status_code in (200, 204), d.text + + # Page 2 via cursor. + r2 = http.get( + api_base + "/api/assets", + params={ + "include_tags": "unit-tests,cursor-delete", + "sort": "name", + "order": "asc", + "limit": "2", + "after": cursor, + }, + timeout=120, + ) + assert r2.status_code == 200, r2.text + body2 = r2.json() + assert [a["name"] for a in body2["assets"]] == names[2:] From 039ed38ed10ad0072a13f6471e06913ed33d5e56 Mon Sep 17 00:00:00 2001 From: Matt Miller Date: Tue, 9 Jun 2026 21:52:14 -0700 Subject: [PATCH 10/21] fix(assets): remove unused delete_content param from deleteAsset (#14241) MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit * fix(assets): remove unused delete_content param from deleteAsset The delete_content query param on DELETE /api/assets/{id} was introduced in #12125 and had its default flipped to false in #12621. In practice no client sends it: the frontend issues a bare DELETE /assets/{id}, so every real caller already gets the default soft-delete (the reference is hidden, content preserved). The only thing that set delete_content=true was this repo's own test teardown. Remove the param from the route and the OpenAPI spec so the contract matches what clients actually use (and lines up with the cloud surface). The route now always soft-deletes. The underlying delete_asset_reference helper keeps its delete_content_if_orphan option, so orphan reclamation remains available internally for a future GC path — it's just no longer exposed on the public endpoint. Tests that used delete_content=true for hard cleanup now soft-delete; test_delete_upon_reference_count asserts content preservation instead of orphan removal. * test/docs: address review on deleteAsset delete_content removal - Rename test_delete_upon_reference_count -> test_soft_delete_preserves_asset_identity_across_references; the old name implied last-ref cleanup, but it now verifies the opposite (soft delete preserves identity across references). - Strengthen the re-association assertion: also check asset_hash == src_hash so it proves content reuse rather than relying on the now-tautological created_new is False. - Document delete_asset_reference: the orphan-reclamation branch is intentionally internal-only; the public endpoint always soft-deletes. - Normalize the soft-delete comment phrasing. * test(assets): make seed content unique per test for isolation Removing the delete_content param means delete is always a soft delete, so content created by one test now survives into the next. The suite had been relying on hard-delete teardown for isolation, so shared fixed-content fixtures started colliding: seeded_asset (b"A"*4096) and make_asset_bytes (deterministic on name) produced the same hash every test, so the second seed deduped to the surviving asset and returned 200 instead of 201, cascading into ~14 failures/errors. Salt both fixtures with a per-test uuid so each test creates fresh content (created_new True, 201), while keeping content deterministic within a test (same name/size -> same bytes) and preserving exact byte length so size-based list/sort assertions are unaffected. --- app/assets/api/routes.py | 10 +++------- app/assets/services/asset_management.py | 10 ++++++++++ tests-unit/assets_test/conftest.py | 21 +++++++++++++++++---- tests-unit/assets_test/test_crud.py | 23 +++++++++++++---------- tests-unit/assets_test/test_downloads.py | 2 +- tests-unit/assets_test/test_tags_api.py | 4 ++-- 6 files changed, 46 insertions(+), 24 deletions(-) diff --git a/app/assets/api/routes.py b/app/assets/api/routes.py index 544a614f2..7ef462f5c 100644 --- a/app/assets/api/routes.py +++ b/app/assets/api/routes.py @@ -533,18 +533,14 @@ async def update_asset_route(request: web.Request) -> web.Response: @_require_assets_feature_enabled async def delete_asset_route(request: web.Request) -> web.Response: reference_id = str(uuid.UUID(request.match_info["id"])) - delete_content_param = request.query.get("delete_content") - delete_content = ( - False - if delete_content_param is None - else delete_content_param.lower() not in {"0", "false", "no"} - ) try: + # Deleting an asset is a soft delete of the reference; the underlying + # content is preserved (it may be shared with other references). deleted = delete_asset_reference( reference_id=reference_id, owner_id=USER_MANAGER.get_request_user_id(request), - delete_content_if_orphan=delete_content, + delete_content_if_orphan=False, ) except Exception: logging.exception( diff --git a/app/assets/services/asset_management.py b/app/assets/services/asset_management.py index 1072c95fa..d4e4fc61c 100644 --- a/app/assets/services/asset_management.py +++ b/app/assets/services/asset_management.py @@ -160,6 +160,16 @@ def delete_asset_reference( owner_id: str, delete_content_if_orphan: bool = True, ) -> bool: + """Delete an asset reference. + + With ``delete_content_if_orphan=False`` (a soft delete), the reference is + hidden and the underlying content is preserved. With ``True``, the content + is also removed once it becomes orphaned. + + Note: the public DELETE /api/assets/{id} endpoint always soft-deletes + (passes ``False``); the orphan-reclamation path is intentionally + internal-only, retained for a future GC/admin caller. + """ with create_session() as session: if not delete_content_if_orphan: # Soft delete: mark the reference as deleted but keep everything diff --git a/tests-unit/assets_test/conftest.py b/tests-unit/assets_test/conftest.py index 9867b4e14..4aa20372f 100644 --- a/tests-unit/assets_test/conftest.py +++ b/tests-unit/assets_test/conftest.py @@ -6,6 +6,7 @@ import subprocess import sys import tempfile import time +import uuid from pathlib import Path from typing import Callable, Iterator, Optional @@ -188,9 +189,17 @@ def _post_multipart_asset( @pytest.fixture def make_asset_bytes() -> Callable[[str, int], bytes]: + # Salt content per test so it never collides with assets left over from + # earlier tests. Delete is now always a soft delete (content is preserved), + # so the suite can no longer rely on hard-deleting content for isolation. + # Deterministic within a test: the same (name, size) yields the same bytes. + salt = uuid.uuid4().bytes + def _make(name: str, size: int = 8192) -> bytes: seed = sum(ord(c) for c in name) % 251 - return bytes((i * 31 + seed) % 256 for i in range(size)) + body = bytearray((i * 31 + seed) % 256 for i in range(size)) + body[: len(salt)] = salt[:size] + return bytes(body) return _make @@ -212,7 +221,7 @@ def asset_factory(http: requests.Session, api_base: str): for aid in created: with contextlib.suppress(Exception): - http.delete(f"{api_base}/api/assets/{aid}?delete_content=true", timeout=30) + http.delete(f"{api_base}/api/assets/{aid}", timeout=30) @pytest.fixture @@ -227,7 +236,11 @@ def seeded_asset(request: pytest.FixtureRequest, http: requests.Session, api_bas if tags is None: tags = ["models", "checkpoints", "unit-tests", "alpha"] meta = {"purpose": "test", "epoch": 1, "flags": ["x", "y"], "nullable": None} - files = {"file": (name, b"A" * 4096, "application/octet-stream")} + # Unique content per test so the seed always creates a fresh asset (201). + # Delete is now always a soft delete, so content from a prior test survives + # and would otherwise dedup this upload into an existing asset (200). + content = uuid.uuid4().bytes + b"A" * (4096 - 16) + files = {"file": (name, content, "application/octet-stream")} form_data = { "tags": json.dumps(tags), "name": name, @@ -260,4 +273,4 @@ def autoclean_unit_test_assets(http: requests.Session, api_base: str): break for aid in ids: with contextlib.suppress(Exception): - http.delete(f"{api_base}/api/assets/{aid}?delete_content=true", timeout=30) + http.delete(f"{api_base}/api/assets/{aid}", timeout=30) diff --git a/tests-unit/assets_test/test_crud.py b/tests-unit/assets_test/test_crud.py index fd2e9a098..36abb60ee 100644 --- a/tests-unit/assets_test/test_crud.py +++ b/tests-unit/assets_test/test_crud.py @@ -45,8 +45,8 @@ def test_get_and_delete_asset(http: requests.Session, api_base: str, seeded_asse assert "user_metadata" in detail assert "filename" in detail["user_metadata"] - # DELETE (hard delete to also remove underlying asset and file) - rd = http.delete(f"{api_base}/api/assets/{aid}?delete_content=true", timeout=120) + # Soft delete — the reference is hidden, content is preserved + rd = http.delete(f"{api_base}/api/assets/{aid}", timeout=120) assert rd.status_code == 204 # GET again -> 404 @@ -60,7 +60,7 @@ def test_soft_delete_hides_from_get(http: requests.Session, api_base: str, seede aid = seeded_asset["id"] asset_hash = seeded_asset["asset_hash"] - # Soft-delete (default, no delete_content param) + # Soft delete — the reference is hidden, content is preserved rd = http.delete(f"{api_base}/api/assets/{aid}", timeout=120) assert rd.status_code == 204 @@ -81,11 +81,10 @@ def test_soft_delete_hides_from_get(http: requests.Session, api_base: str, seede ids = [a["id"] for a in rl.json().get("assets", [])] assert aid not in ids - # Clean up: hard-delete the soft-deleted reference and orphaned asset - http.delete(f"{api_base}/api/assets/{aid}?delete_content=true", timeout=120) + # The reference is already soft-deleted; content is preserved. -def test_delete_upon_reference_count( +def test_soft_delete_preserves_asset_identity_across_references( http: requests.Session, api_base: str, seeded_asset: dict ): # Create a second reference to the same asset via from-hash @@ -119,16 +118,20 @@ def test_delete_upon_reference_count( rh2 = http.head(f"{api_base}/api/assets/hash/{src_hash}", timeout=120) assert rh2.status_code == 200 # asset identity preserved (soft delete) - # Re-associate via from-hash, then hard-delete -> orphan content removed + # Re-associate via from-hash: it must reuse the same preserved content + # (created_new False AND the same hash), proving the soft deletes did not + # destroy the underlying asset. Then soft-delete again -> still preserved. r3 = http.post(f"{api_base}/api/assets/from-hash", json=payload, timeout=120) assert r3.status_code == 201, r3.json() + assert r3.json()["created_new"] is False + assert r3.json()["asset_hash"] == src_hash # reused the surviving content aid3 = r3.json()["id"] - rd3 = http.delete(f"{api_base}/api/assets/{aid3}?delete_content=true", timeout=120) + rd3 = http.delete(f"{api_base}/api/assets/{aid3}", timeout=120) assert rd3.status_code == 204 rh3 = http.head(f"{api_base}/api/assets/hash/{src_hash}", timeout=120) - assert rh3.status_code == 404 # orphan content removed + assert rh3.status_code == 200 # content preserved (soft delete) def test_update_asset_fields(http: requests.Session, api_base: str, seeded_asset: dict): @@ -249,7 +252,7 @@ def test_concurrent_delete_same_asset_info_single_204( # Hit the same endpoint N times in parallel. n_tests = 4 - url = f"{api_base}/api/assets/{aid}?delete_content=false" + url = f"{api_base}/api/assets/{aid}" def _do_delete(delete_url): with requests.Session() as s: diff --git a/tests-unit/assets_test/test_downloads.py b/tests-unit/assets_test/test_downloads.py index 672ba9728..42c64a5fd 100644 --- a/tests-unit/assets_test/test_downloads.py +++ b/tests-unit/assets_test/test_downloads.py @@ -117,7 +117,7 @@ def test_download_missing_file_returns_404( assert body["error"]["code"] == "FILE_NOT_FOUND" finally: # We created asset without the "unit-tests" tag(see `autoclean_unit_test_assets`), we need to clear it manually. - dr = http.delete(f"{api_base}/api/assets/{aid}?delete_content=true", timeout=120) + dr = http.delete(f"{api_base}/api/assets/{aid}", timeout=120) dr.content diff --git a/tests-unit/assets_test/test_tags_api.py b/tests-unit/assets_test/test_tags_api.py index 595bf29c6..9729b7d03 100644 --- a/tests-unit/assets_test/test_tags_api.py +++ b/tests-unit/assets_test/test_tags_api.py @@ -69,8 +69,8 @@ def test_tags_empty_usage(http: requests.Session, api_base: str, asset_factory, used_names = [t["name"] for t in body2["tags"]] assert custom_tag in used_names - # Hard-delete the asset so the tag usage drops to zero - rd = http.delete(f"{api_base}/api/assets/{_asset['id']}?delete_content=true", timeout=120) + # Delete the asset reference so the tag usage drops to zero + rd = http.delete(f"{api_base}/api/assets/{_asset['id']}", timeout=120) assert rd.status_code == 204 # Now the custom tag must not be returned when include_zero=false From 6d18f4adacea2304f1f6f4ff3c0279d13654ec5c Mon Sep 17 00:00:00 2001 From: rattus <46076784+rattus128@users.noreply.github.com> Date: Thu, 11 Jun 2026 03:54:32 +1000 Subject: [PATCH 11/21] main: force cudnn.benchmark to false (#14390) Some custom nodes try to set this true globally. It messes with dynamic VRAM with one-off spikes that can OOM but this is also very high risk for windows where such allocations might get serviced by shared memory fallback. Trump it. --- comfy/model_management.py | 6 ++++-- main.py | 5 +++++ 2 files changed, 9 insertions(+), 2 deletions(-) diff --git a/comfy/model_management.py b/comfy/model_management.py index 9dc0a4e13..55ddaab8e 100644 --- a/comfy/model_management.py +++ b/comfy/model_management.py @@ -534,8 +534,10 @@ try: except: pass -if torch.cuda.is_available() and torch.backends.cudnn.is_available() and PerformanceFeature.AutoTune in args.fast: - torch.backends.cudnn.benchmark = True + +def set_cudnn_benchmark(): + if torch.cuda.is_available() and torch.backends.cudnn.is_available(): + torch.backends.cudnn.benchmark = PerformanceFeature.AutoTune in args.fast try: if torch_version_numeric >= (2, 5): diff --git a/main.py b/main.py index 7fcc8e97d..0ad660376 100644 --- a/main.py +++ b/main.py @@ -490,6 +490,11 @@ def start_comfyui(asyncio_loop=None): init_custom_nodes=(not args.disable_all_custom_nodes) or len(args.whitelist_custom_nodes) > 0, init_api_nodes=not args.disable_api_nodes )) + + # Re-apply Comfy's cuDNN benchmark policy after custom-node imports. Benchmark + # mode can request near-card-sized autotune workspaces, and some custom nodes set it at import time. + comfy.model_management.set_cudnn_benchmark() + hook_breaker_ac10a0.restore_functions() cuda_malloc_warning() From e5b7140dcc5a88a6ad673a249eed223238e45a2b Mon Sep 17 00:00:00 2001 From: Matt Miller Date: Wed, 10 Jun 2026 16:55:25 -0700 Subject: [PATCH 12/21] feat(assets): add job_ids filter to GET /api/assets (#13998) MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit * feat(assets): add job_ids filter to GET /api/assets Mirrors the existing cloud `job_ids` query param on the local Python server: clients can pass a comma-separated list (or repeated query params) of UUIDs to filter assets by their associated job. The `AssetReference.job_id` column already exists, so no migration is needed — this just plumbs the filter through schema → service → query. Marks the parameter as available in both runtimes by dropping the `[cloud-only]` description prefix and the `x-runtime: [cloud]` tag from the OpenAPI spec, per the OSS field-drift convention (absent runtime tag = populated by both local and cloud). * fix(assets): tighten job_ids — array schema, max_length, narrow except From cursor-reviews on the parent commit: - OpenAPI: declare job_ids as `type: array, items: string format: uuid` with `style: form, explode: true` so it matches the documented contract (and matches sibling include_tags/exclude_tags shape). Description now states both accepted shapes explicitly. - Schema: cap `job_ids` at 500 entries (max_length on the Pydantic field) so a client can't splice an unbounded list into the IN clauses. - Schema: drop `AttributeError` from the except — `raw` only contains `str` items by construction, so `uuid.UUID()` raises `ValueError` exclusively; the second clause was dead code. * fix(assets): tighten job_ids validator + add schema-level tests Aligns with the parallel hardening from draft PR #13848 (now closed as a duplicate). The validator now: - Raises ValueError on non-string list items (was: silently dropped). - Raises ValueError on non-string / non-list top-level values like dict or int (was: silently passed through to Pydantic's downstream coercion). Adds tests-unit/assets_test/queries/test_list_assets_query.py covering the validator end-to-end: CSV canonicalization, dedup order, default empty, invalid UUID, non-string list item, non-string non-list value, and the max_length=500 boundary. * feat(prompt): enforce canonical UUID prompt_id at job creation POST /prompt previously accepted any client-supplied prompt_id verbatim, str()-coercing even non-strings, and minting the literal job id "None" for an explicit JSON null. The new GET /api/assets job_ids filter matches stored job ids as canonical UUIDs exactly, so a non-UUID id minted a job whose assets could never be filtered. - validate_job_id (comfy_execution/jobs.py): requires a string in the canonical lowercase hyphenated UUID form; raises ValueError otherwise, including parseable-but-non-canonical spellings (uppercase, braced, URN, bare hex), which would otherwise be silently rewritten and then miss every exact-match lookup downstream (history keys, websocket correlation, /interrupt, the assets job_ids filter). - POST /prompt: absent or null prompt_id means the server mints uuid4; invalid means 400 invalid_prompt_id on the standard error envelope. - openapi.yaml: document the request-side prompt_id (format uuid, nullable) on PromptRequest. - tests: unit matrix for validate_job_id; integration tests against the booted server covering rejection, acceptance, and null handling. --------- Co-authored-by: guill --- app/assets/api/routes.py | 1 + app/assets/api/schemas_in.py | 36 ++++++++++ .../database/queries/asset_reference.py | 6 ++ app/assets/services/asset_management.py | 2 + comfy_execution/jobs.py | 21 ++++++ openapi.yaml | 5 ++ server.py | 18 ++++- .../assets_test/queries/test_asset_info.py | 50 ++++++++++++++ .../queries/test_list_assets_query.py | 60 ++++++++++++++++ .../assets_test/test_prompt_id_enforcement.py | 69 +++++++++++++++++++ tests/execution/test_jobs.py | 43 ++++++++++++ 11 files changed, 309 insertions(+), 2 deletions(-) create mode 100644 tests-unit/assets_test/queries/test_list_assets_query.py create mode 100644 tests-unit/assets_test/test_prompt_id_enforcement.py diff --git a/app/assets/api/routes.py b/app/assets/api/routes.py index 7ef462f5c..6c9a3200d 100644 --- a/app/assets/api/routes.py +++ b/app/assets/api/routes.py @@ -219,6 +219,7 @@ async def list_assets_route(request: web.Request) -> web.Response: exclude_tags=q.exclude_tags, name_contains=q.name_contains, metadata_filter=q.metadata_filter, + job_ids=q.job_ids, limit=q.limit, offset=q.offset, sort=sort, diff --git a/app/assets/api/schemas_in.py b/app/assets/api/schemas_in.py index af666746d..4ae18c65a 100644 --- a/app/assets/api/schemas_in.py +++ b/app/assets/api/schemas_in.py @@ -1,4 +1,5 @@ import json +import uuid from dataclasses import dataclass from typing import Any, Literal @@ -53,6 +54,7 @@ class ListAssetsQuery(BaseModel): include_tags: list[str] = Field(default_factory=list) exclude_tags: list[str] = Field(default_factory=list) name_contains: str | None = None + job_ids: list[str] = Field(default_factory=list, max_length=500) # Accept either a JSON string (query param) or a dict metadata_filter: dict[str, Any] | None = None @@ -86,6 +88,40 @@ class ListAssetsQuery(BaseModel): return out return v + @field_validator("job_ids", mode="before") + @classmethod + def _split_and_validate_job_ids(cls, v): + # Accept "uuid1,uuid2" or ["uuid1","uuid2"] or repeated query params. + # Each entry must parse as a UUID; canonicalized to lowercase hyphenated form. + if v is None: + return [] + if isinstance(v, str): + raw = [t.strip() for t in v.split(",") if t.strip()] + elif isinstance(v, list): + raw = [] + for item in v: + if not isinstance(item, str): + raise ValueError( + f"job_ids entries must be strings, got {type(item).__name__}" + ) + raw.extend([t.strip() for t in item.split(",") if t.strip()]) + else: + raise ValueError( + f"job_ids must be a string or list of strings, got {type(v).__name__}" + ) + + out: list[str] = [] + seen: set[str] = set() + for s in raw: + try: + canonical = str(uuid.UUID(s)) + except ValueError as e: + raise ValueError(f"job_ids must be UUIDs: {s!r}") from e + if canonical not in seen: + seen.add(canonical) + out.append(canonical) + return out + @field_validator("metadata_filter", mode="before") @classmethod def _parse_metadata_json(cls, v): diff --git a/app/assets/database/queries/asset_reference.py b/app/assets/database/queries/asset_reference.py index 792411800..33ded8a1c 100644 --- a/app/assets/database/queries/asset_reference.py +++ b/app/assets/database/queries/asset_reference.py @@ -264,6 +264,7 @@ def list_references_page( include_tags: Sequence[str] | None = None, exclude_tags: Sequence[str] | None = None, metadata_filter: dict | None = None, + job_ids: Sequence[str] | None = None, sort: str | None = None, order: str | None = None, after_cursor_value: object | None = None, @@ -293,6 +294,9 @@ def list_references_page( escaped, esc = escape_sql_like_string(name_contains) base = base.where(AssetReference.name.ilike(f"%{escaped}%", escape=esc)) + if job_ids: + base = base.where(AssetReference.job_id.in_(list(job_ids))) + base = apply_tag_filters(base, include_tags, exclude_tags) base = apply_metadata_filter(base, metadata_filter) @@ -345,6 +349,8 @@ def list_references_page( count_stmt = count_stmt.where( AssetReference.name.ilike(f"%{escaped}%", escape=esc) ) + if job_ids: + count_stmt = count_stmt.where(AssetReference.job_id.in_(list(job_ids))) count_stmt = apply_tag_filters(count_stmt, include_tags, exclude_tags) count_stmt = apply_metadata_filter(count_stmt, metadata_filter) diff --git a/app/assets/services/asset_management.py b/app/assets/services/asset_management.py index d4e4fc61c..53aec7a15 100644 --- a/app/assets/services/asset_management.py +++ b/app/assets/services/asset_management.py @@ -274,6 +274,7 @@ def list_assets_page( exclude_tags: Sequence[str] | None = None, name_contains: str | None = None, metadata_filter: dict | None = None, + job_ids: Sequence[str] | None = None, limit: int = 20, offset: int = 0, sort: str = "created_at", @@ -319,6 +320,7 @@ def list_assets_page( exclude_tags=exclude_tags, name_contains=name_contains, metadata_filter=metadata_filter, + job_ids=job_ids, limit=fetch_limit, offset=offset, sort=sort, diff --git a/comfy_execution/jobs.py b/comfy_execution/jobs.py index fcd7ef735..3fbcc3eb0 100644 --- a/comfy_execution/jobs.py +++ b/comfy_execution/jobs.py @@ -3,6 +3,7 @@ Job utilities for the /api/jobs endpoint. Provides normalization and helper functions for job status tracking. """ +import uuid from typing import Optional from comfy_api.internal import prune_dict @@ -19,6 +20,26 @@ class JobStatus: ALL = [PENDING, IN_PROGRESS, COMPLETED, FAILED, CANCELLED] +def validate_job_id(value) -> str: + """Validate a client-supplied job (prompt) id. + + Job ids must be UUIDs in the canonical lowercase hyphenated form. The id + is stored and compared verbatim everywhere downstream — history keys, + websocket events, /interrupt matching, and the assets ``job_ids`` filter + (a String(36) column matched exactly) — so accepting another spelling + would either rewrite the client's id behind its back or mint a job whose + outputs the filter can never find. Rejecting loudly beats both. + + Returns the id unchanged. Raises ValueError when the value is not a + string in canonical UUID form. + """ + if not isinstance(value, str): + raise ValueError(f"job id must be a string, got {type(value).__name__}") + if str(uuid.UUID(value)) != value: + raise ValueError("job id must be a UUID in canonical lowercase hyphenated form") + return value + + # Media types that can be previewed in the frontend PREVIEWABLE_MEDIA_TYPES = frozenset({'images', 'video', 'audio', '3d', 'text'}) diff --git a/openapi.yaml b/openapi.yaml index c27ed7adf..58614103a 100644 --- a/openapi.yaml +++ b/openapi.yaml @@ -896,6 +896,11 @@ components: additionalProperties: true description: The workflow graph to execute type: object + prompt_id: + description: Optional client-supplied job id. Must be a UUID in canonical lowercase hyphenated form; it is echoed back in the response. Omitted or null means the server generates one. + format: uuid + nullable: true + type: string workflow_id: description: UUID identifying the cloud workflow entity to associate with this job type: string diff --git a/server.py b/server.py index a85c1e591..cc3b33a5c 100644 --- a/server.py +++ b/server.py @@ -8,7 +8,7 @@ import time import nodes import folder_paths import execution -from comfy_execution.jobs import JobStatus, get_job, get_all_jobs +from comfy_execution.jobs import JobStatus, get_job, get_all_jobs, validate_job_id import uuid import urllib import json @@ -942,7 +942,21 @@ class PromptServer(): if "prompt" in json_data: prompt = json_data["prompt"] - prompt_id = str(json_data.get("prompt_id", uuid.uuid4())) + client_prompt_id = json_data.get("prompt_id") + if client_prompt_id is None: + # Absent or explicit null: the server mints the id. + prompt_id = str(uuid.uuid4()) + else: + try: + prompt_id = validate_job_id(client_prompt_id) + except ValueError: + error = { + "type": "invalid_prompt_id", + "message": "prompt_id must be a valid UUID", + "details": "prompt_id must be a UUID string in canonical lowercase hyphenated form; omit it to let the server generate one", + "extra_info": {} + } + return web.json_response({"error": error, "node_errors": {}}, status=400) partial_execution_targets = None if "partial_execution_targets" in json_data: diff --git a/tests-unit/assets_test/queries/test_asset_info.py b/tests-unit/assets_test/queries/test_asset_info.py index fe510e342..ba729a270 100644 --- a/tests-unit/assets_test/queries/test_asset_info.py +++ b/tests-unit/assets_test/queries/test_asset_info.py @@ -158,6 +158,56 @@ class TestListReferencesPage: refs, _, _ = list_references_page(session, sort="name", order="asc") assert refs[0].name == "large" + def test_job_ids_filter(self, session: Session): + asset = _make_asset(session, "hash1") + job_a = str(uuid.uuid4()) + job_b = str(uuid.uuid4()) + ref_a = _make_reference(session, asset, name="from_job_a") + ref_a.job_id = job_a + ref_b = _make_reference(session, asset, name="from_job_b") + ref_b.job_id = job_b + _make_reference(session, asset, name="no_job") + session.commit() + + # Single job filter + refs, _, total = list_references_page(session, job_ids=[job_a]) + assert total == 1 + assert refs[0].name == "from_job_a" + + # Multi-job filter (IN) + refs, _, total = list_references_page(session, job_ids=[job_a, job_b]) + names = sorted(r.name for r in refs) + assert total == 2 + assert names == ["from_job_a", "from_job_b"] + + # Unknown job id matches nothing + refs, _, total = list_references_page(session, job_ids=[str(uuid.uuid4())]) + assert total == 0 + assert refs == [] + + # Empty/None means no filter -> all three references + refs, _, total = list_references_page(session, job_ids=[]) + assert total == 3 + refs, _, total = list_references_page(session, job_ids=None) + assert total == 3 + + def test_job_ids_combined_with_other_filters(self, session: Session): + asset = _make_asset(session, "hash1") + job_a = str(uuid.uuid4()) + ref_match = _make_reference(session, asset, name="match.bin") + ref_match.job_id = job_a + ref_wrong_name = _make_reference(session, asset, name="other.bin") + ref_wrong_name.job_id = job_a + ref_wrong_job = _make_reference(session, asset, name="match.bin") + ref_wrong_job.job_id = str(uuid.uuid4()) + session.commit() + + refs, _, total = list_references_page( + session, job_ids=[job_a], name_contains="match" + ) + assert total == 1 + assert refs[0].id == ref_match.id + class TestFetchReferenceAssetAndTags: def test_returns_none_for_nonexistent(self, session: Session): diff --git a/tests-unit/assets_test/queries/test_list_assets_query.py b/tests-unit/assets_test/queries/test_list_assets_query.py new file mode 100644 index 000000000..e8d3430e2 --- /dev/null +++ b/tests-unit/assets_test/queries/test_list_assets_query.py @@ -0,0 +1,60 @@ +"""Schema-level unit tests for ListAssetsQuery (no DB required).""" +import uuid + +import pytest +from pydantic import ValidationError + +from app.assets.api.schemas_in import ListAssetsQuery + + +class TestJobIdsValidator: + def test_csv_string_parses_and_canonicalizes(self): + a = "AAAAAAAA-BBBB-CCCC-DDDD-EEEEEEEEEEEE" + b = "11111111-2222-3333-4444-555555555555" + q = ListAssetsQuery.model_validate({"job_ids": f"{a},{b}"}) + # Canonicalized to lowercase + assert q.job_ids == [a.lower(), b] + + def test_repeated_query_params_as_list(self): + a = "11111111-1111-1111-1111-111111111111" + b = "22222222-2222-2222-2222-222222222222" + q = ListAssetsQuery.model_validate({"job_ids": [a, b]}) + assert q.job_ids == [a, b] + + def test_dedup_preserves_first_seen_order(self): + a = "11111111-1111-1111-1111-111111111111" + b = "22222222-2222-2222-2222-222222222222" + q = ListAssetsQuery.model_validate({"job_ids": [a, b, a]}) + assert q.job_ids == [a, b] + + def test_default_empty(self): + q = ListAssetsQuery.model_validate({}) + assert q.job_ids == [] + + def test_invalid_uuid_rejected(self): + with pytest.raises(ValidationError) as exc: + ListAssetsQuery.model_validate({"job_ids": "not-a-uuid"}) + assert "must be UUIDs" in str(exc.value) + + def test_non_string_list_item_rejected(self): + with pytest.raises(ValidationError) as exc: + ListAssetsQuery.model_validate( + {"job_ids": ["11111111-1111-1111-1111-111111111111", 42]} + ) + assert "must be strings" in str(exc.value) + + def test_non_string_non_list_value_rejected(self): + with pytest.raises(ValidationError) as exc: + ListAssetsQuery.model_validate({"job_ids": {"bad": "shape"}}) + assert "must be a string or list of strings" in str(exc.value) + + def test_max_length_enforced(self): + too_many = [str(uuid.uuid4()) for _ in range(501)] + with pytest.raises(ValidationError) as exc: + ListAssetsQuery.model_validate({"job_ids": too_many}) + assert exc.value.errors()[0]["type"] == "too_long" + + def test_max_length_boundary_accepted(self): + at_cap = [str(uuid.uuid4()) for _ in range(500)] + q = ListAssetsQuery.model_validate({"job_ids": at_cap}) + assert len(q.job_ids) == 500 diff --git a/tests-unit/assets_test/test_prompt_id_enforcement.py b/tests-unit/assets_test/test_prompt_id_enforcement.py new file mode 100644 index 000000000..fb961beae --- /dev/null +++ b/tests-unit/assets_test/test_prompt_id_enforcement.py @@ -0,0 +1,69 @@ +"""POST /prompt enforces canonical-UUID job ids at creation time. + +Lives in assets_test because it uses this suite's booted-server fixture and +because the invariant exists for the assets pipeline: the GET /api/assets +``job_ids`` filter matches stored job ids exactly, so a job minted with a +non-canonical id would produce assets the filter can never find. + +The prompt bodies here are intentionally invalid workflows — prompt_id +validation happens before workflow validation, so a rejected id returns +``invalid_prompt_id`` while an accepted id falls through to the ordinary +workflow-validation error (proving it cleared the id check). +""" +import requests + + +def _post_prompt(http: requests.Session, api_base: str, body: dict) -> requests.Response: + return http.post(api_base + "/prompt", json=body, timeout=30) + + +def _error_type(r: requests.Response) -> str: + return r.json()["error"]["type"] + + +def test_non_uuid_prompt_id_rejected(http: requests.Session, api_base: str): + r = _post_prompt(http, api_base, {"prompt": {}, "prompt_id": "not-a-uuid"}) + assert r.status_code == 400, r.text + assert _error_type(r) == "invalid_prompt_id" + + +def test_non_string_prompt_id_rejected(http: requests.Session, api_base: str): + # Previously str()-coerced (123 became the job id "123"); must now be a 400, + # not a 500 from uuid.UUID choking on a non-string. + r = _post_prompt(http, api_base, {"prompt": {}, "prompt_id": 123}) + assert r.status_code == 400, r.text + assert _error_type(r) == "invalid_prompt_id" + + +def test_non_canonical_uuid_rejected(http: requests.Session, api_base: str): + # Parseable as a UUID, but not the canonical lowercase form: rejected + # loudly rather than silently rewritten (downstream lookups match the + # stored id exactly). + r = _post_prompt( + http, + api_base, + {"prompt": {}, "prompt_id": "AAAAAAAA-BBBB-4CCC-8DDD-EEEEEEEEEEEE"}, + ) + assert r.status_code == 400, r.text + assert _error_type(r) == "invalid_prompt_id" + + +def test_canonical_uuid_accepted(http: requests.Session, api_base: str): + # The id clears validation; the empty workflow then fails ordinary prompt + # validation, proving the request got past the id check. + r = _post_prompt( + http, + api_base, + {"prompt": {}, "prompt_id": "aaaaaaaa-bbbb-4ccc-8ddd-eeeeeeeeeeee"}, + ) + assert r.status_code == 400, r.text + assert _error_type(r) != "invalid_prompt_id" + + +def test_null_prompt_id_not_rejected(http: requests.Session, api_base: str): + # Explicit null means "server generates" and must not be rejected as an + # invalid id. (The minted id itself is not observable here because the + # workflow is invalid; unit tests cover validate_job_id directly.) + r = _post_prompt(http, api_base, {"prompt": {}, "prompt_id": None}) + assert r.status_code == 400, r.text + assert _error_type(r) != "invalid_prompt_id" diff --git a/tests/execution/test_jobs.py b/tests/execution/test_jobs.py index 814af5c13..30e47071d 100644 --- a/tests/execution/test_jobs.py +++ b/tests/execution/test_jobs.py @@ -1,5 +1,7 @@ """Unit tests for comfy_execution/jobs.py""" +import pytest + from comfy_execution.jobs import ( JobStatus, is_previewable, @@ -10,9 +12,50 @@ from comfy_execution.jobs import ( get_outputs_summary, apply_sorting, has_3d_extension, + validate_job_id, ) +class TestValidateJobId: + """validate_job_id guards job creation: POST /prompt rejects ids it raises on.""" + + def test_canonical_form_passes_through(self): + cid = "a1b2c3d4-e5f6-7a89-b0c1-d2e3f4a5b6c7" + assert validate_job_id(cid) == cid + + @pytest.mark.parametrize( + "variant", + [ + "A1B2C3D4-E5F6-7A89-B0C1-D2E3F4A5B6C7", # uppercase + "{a1b2c3d4-e5f6-7a89-b0c1-d2e3f4a5b6c7}", # braced + "urn:uuid:a1b2c3d4-e5f6-7a89-b0c1-d2e3f4a5b6c7", # URN + "a1b2c3d4e5f67a89b0c1d2e3f4a5b6c7", # bare hex + " a1b2c3d4-e5f6-7a89-b0c1-d2e3f4a5b6c7 ", # padded + ], + ) + def test_non_canonical_spellings_rejected(self, variant): + # uuid.UUID parses all of these, but accepting them would silently + # rewrite the client's id (history keys, websocket events, and the + # assets job_ids filter all match the stored form exactly). + with pytest.raises(ValueError): + validate_job_id(variant) + + @pytest.mark.parametrize( + "bad", + ["", "not-a-uuid", "prompt-123", "a1b2c3d4-e5f6-7a89-b0c1", "None"], + ) + def test_non_uuid_strings_rejected(self, bad): + with pytest.raises(ValueError): + validate_job_id(bad) + + @pytest.mark.parametrize("bad", [123, 1.5, True, None, ["a"], {"id": "x"}]) + def test_non_strings_rejected(self, bad): + # uuid.UUID raises AttributeError/TypeError on non-strings; the helper + # must normalize those to ValueError so callers need one except clause. + with pytest.raises(ValueError): + validate_job_id(bad) + + class TestJobStatus: """Test JobStatus constants.""" From ce200c0850182722cfd6e0f9f9bd3f619e48281e Mon Sep 17 00:00:00 2001 From: Matt Miller Date: Wed, 10 Jun 2026 17:04:52 -0700 Subject: [PATCH 13/21] feat(assets): include asset id in executed WebSocket message (#13862) MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit * feat(assets): enrich executed WS message with asset metadata When --enable-assets is set, each file-type output entry in the `executed` WebSocket message now includes id, name, asset_hash, size, and mime_type — matching the shape already returned by /upload/image. The enrichment lives in comfy_execution/asset_enrichment.py (no torch dependency) and is called from both send sites in execution.py: freshly executed nodes register the file inline via register_file_in_place; cached node re-sends look up the existing AssetReference by file path to avoid re-hashing. Errors are caught per-entry so a failure never blocks the WS message from sending. * fix(assets): inject only id in executed WS message per Asset Identity RFC Per the Asset Identity RFC, the executed WebSocket payload should carry id alone — hash is already encoded in the filename, and name/preview_url/ size belong behind GET /api/assets/{id} rather than being pushed eagerly. Simplifies the DB lookup path: we only need ref.id, so the asset.hash null-check is no longer required as a fallback trigger. * fix(assets): reject path traversal when resolving output abs_path Subfolder/filename were joined and absolutized without containment check, so '..' segments or an absolute filename could escape the type's base directory and register an unrelated on-disk file as an asset. Add commonpath-based containment check; skip enrichment (warn, leave entry unchanged) when the resolved path escapes base. Catches ValueError from cross-drive paths on Windows. * docs(assets): drop Asset Identity RFC reference from docstring * docs(assets): trim docstring to what enrichment does, not what it doesn't * test(assets): use real platform paths so containment check works on Windows The previous test setup patched os.path.abspath to identity and used a POSIX-style '/output' base, which collided with Windows path separators in os.path.commonpath. Drop the abspath/join patches and use a real tempdir-rooted base so the containment check runs against actual platform paths. * refactor(assets): enrich at output-processing time, not in the WS send path Per review: enrichment lived inside the client_id-guarded send sites, so a headless run (no websocket client) never registered assets at all, and ui_outputs/history stored the un-enriched entries. Now output_ui is enriched once, right after the node produces it and before it is stored in ui_outputs — so registration happens regardless of connected clients, and the asset id flows into history and the execution cache for free. _send_cached_ui re-sends the stored (already-enriched) dict verbatim, which lets the DB-lookup-by-path fallback be deleted: every enrichment is now a fresh output, and register_file_in_place re-hashes on upsert so an overwritten path can never carry a stale id. --- comfy_execution/asset_enrichment.py | 66 ++++++ execution.py | 6 + .../execution_test/test_enrich_output.py | 205 ++++++++++++++++++ 3 files changed, 277 insertions(+) create mode 100644 comfy_execution/asset_enrichment.py create mode 100644 tests-unit/execution_test/test_enrich_output.py diff --git a/comfy_execution/asset_enrichment.py b/comfy_execution/asset_enrichment.py new file mode 100644 index 000000000..38e9496a8 --- /dev/null +++ b/comfy_execution/asset_enrichment.py @@ -0,0 +1,66 @@ +"""Enrich executed-node output entries with asset id.""" +import logging +import os + + +def enrich_output_with_assets(output_ui: dict) -> dict: + """Register file-type output entries as assets and inject their ``id``. + + Runs at output-processing time, once per produced output, when + --enable-assets is set. Returns a new dict; entries without a resolvable + on-disk file path are left unchanged. Errors are caught per-entry so a + failure never blocks execution or the other entries. + """ + from comfy.cli_args import args + if not args.enable_assets: + return output_ui + + import folder_paths + from app.assets.services.ingest import register_file_in_place, DependencyMissingError + + enriched = {} + for key, entries in output_ui.items(): + if not isinstance(entries, list): + enriched[key] = entries + continue + new_entries = [] + for entry in entries: + if not isinstance(entry, dict) or "filename" not in entry or "type" not in entry: + new_entries.append(entry) + continue + try: + base = folder_paths.get_directory_by_type(entry["type"]) + if base is None: + new_entries.append(entry) + continue + base_abs = os.path.abspath(base) + abs_path = os.path.abspath(os.path.join(base_abs, entry.get("subfolder") or "", entry["filename"])) + try: + if os.path.commonpath([base_abs, abs_path]) != base_abs: + raise ValueError("escapes base") + except ValueError: + logging.warning("Asset enrichment skipped (path escapes base): %s", entry.get("filename")) + new_entries.append(entry) + continue + if not os.path.isfile(abs_path): + new_entries.append(entry) + continue + + # Register unconditionally: the file was just produced, and + # register_file_in_place re-hashes so an overwritten path can + # never carry a stale id. + result = register_file_in_place( + abs_path=abs_path, + name=entry["filename"], + tags=[entry["type"]], + ) + + entry = dict(entry) + entry["id"] = result.ref.id + except DependencyMissingError: + logging.warning("Asset enrichment skipped (blake3 not available): %s", entry.get("filename")) + except Exception: + logging.warning("Failed to enrich output entry with asset id: %s", entry.get("filename"), exc_info=True) + new_entries.append(entry) + enriched[key] = new_entries + return enriched diff --git a/execution.py b/execution.py index 5246d651c..e6c6f39d6 100644 --- a/execution.py +++ b/execution.py @@ -40,6 +40,7 @@ from comfy_execution.graph_utils import GraphBuilder, is_link from comfy_execution.validation import validate_node_input from comfy_execution.progress import get_progress_state, reset_progress_state, add_progress_handler, WebUIProgressHandler from comfy_execution.utils import CurrentNodeContext +from comfy_execution.asset_enrichment import enrich_output_with_assets from comfy_api.internal import _ComfyNodeInternal, _NodeOutputInternal, first_real_override, is_class, make_locked_method_func from comfy_api.latest import io, _io from comfy_execution.cache_provider import _has_cache_providers, _get_cache_providers, _logger as _cache_logger @@ -418,6 +419,7 @@ def _is_intermediate_output(dynprompt, node_id): class_def = nodes.NODE_CLASS_MAPPINGS[class_type] return getattr(class_def, 'HAS_INTERMEDIATE_OUTPUT', False) + def _send_cached_ui(server, node_id, display_node_id, cached, prompt_id, ui_outputs): if server.client_id is None: return @@ -552,6 +554,10 @@ async def execute(server, dynprompt, caches, current_item, extra_data, executed, asyncio.create_task(await_completion()) return (ExecutionResult.PENDING, None, None) if len(output_ui) > 0: + # Enrich at output-processing time (not in the send path) so assets + # are registered even when no client is connected, and the asset id + # flows into ui_outputs and the cache alongside the raw entries. + output_ui = enrich_output_with_assets(output_ui) ui_outputs[unique_id] = { "meta": { "node_id": unique_id, diff --git a/tests-unit/execution_test/test_enrich_output.py b/tests-unit/execution_test/test_enrich_output.py new file mode 100644 index 000000000..61490c49e --- /dev/null +++ b/tests-unit/execution_test/test_enrich_output.py @@ -0,0 +1,205 @@ +"""Tests for enrich_output_with_assets in comfy_execution/asset_enrichment.py.""" +import os +import types +import unittest +from unittest.mock import MagicMock, patch + + +def _make_args(enable_assets: bool): + a = types.SimpleNamespace() + a.enable_assets = enable_assets + return a + + +def _make_register_result(ref_id="ref-id-2"): + result = MagicMock() + result.ref.id = ref_id + return result + + +# Platform-appropriate absolute base. tempfile.gettempdir() returns C:\... on +# Windows and /tmp on POSIX, so containment via commonpath behaves naturally. +_DEFAULT_BASE = os.path.join(__import__("tempfile").gettempdir(), "asset-enrichment-test-base") + + +def _mocked_modules(*, enable_assets=True, register_file_in_place=None, directory=_DEFAULT_BASE): + return { + "comfy.cli_args": MagicMock(args=_make_args(enable_assets)), + "folder_paths": MagicMock(get_directory_by_type=MagicMock(return_value=directory)), + "app.assets.services.ingest": MagicMock( + register_file_in_place=register_file_in_place or MagicMock(return_value=_make_register_result()), + DependencyMissingError=type("DependencyMissingError", (Exception,), {}), + ), + } + + +def _call(output_ui, *, enable_assets=True, file_exists=True, register_result=None, directory=_DEFAULT_BASE): + register_mock = MagicMock(return_value=register_result or _make_register_result()) + mocked = _mocked_modules( + enable_assets=enable_assets, + register_file_in_place=register_mock, + directory=directory, + ) + + # Only os.path.isfile is patched — abspath/join must run natively so the + # containment check sees real platform paths. + with patch.dict("sys.modules", mocked), \ + patch("os.path.isfile", return_value=file_exists): + import importlib + import comfy_execution.asset_enrichment as mod + importlib.reload(mod) + return mod.enrich_output_with_assets(output_ui) + + +class TestEnrichOutputWithAssets(unittest.TestCase): + + def test_disabled_returns_unchanged(self): + output = {"images": [{"filename": "a.png", "subfolder": "", "type": "output"}]} + result = _call(output, enable_assets=False) + self.assertNotIn("id", result["images"][0]) + + def test_non_list_value_passed_through(self): + output = {"text": "hello"} + result = _call(output) + self.assertEqual(result["text"], "hello") + + def test_entry_without_filename_unchanged(self): + output = {"latent": [{"subfolder": "", "type": "output"}]} + result = _call(output) + self.assertNotIn("id", result["latent"][0]) + + def test_entry_without_type_unchanged(self): + output = {"data": [{"filename": "a.png", "subfolder": ""}]} + result = _call(output) + self.assertNotIn("id", result["data"][0]) + + def test_file_not_on_disk_unchanged(self): + output = {"images": [{"filename": "missing.png", "subfolder": "", "type": "output"}]} + result = _call(output, file_exists=False) + self.assertNotIn("id", result["images"][0]) + + def test_unknown_type_returns_none_directory_unchanged(self): + output = {"images": [{"filename": "a.png", "subfolder": "", "type": "unknown"}]} + result = _call(output, directory=None) + self.assertNotIn("id", result["images"][0]) + + def test_register_injects_only_id(self): + reg = _make_register_result(ref_id="inline-ref") + output = {"images": [{"filename": "new.png", "subfolder": "", "type": "output"}]} + result = _call(output, register_result=reg) + img = result["images"][0] + self.assertEqual(img["id"], "inline-ref") + # Only id is injected — no asset_hash, name, preview_url, size + self.assertNotIn("asset_hash", img) + self.assertNotIn("name", img) + self.assertNotIn("preview_url", img) + self.assertNotIn("size", img) + + def test_register_called_per_entry(self): + register_mock = MagicMock(return_value=_make_register_result()) + mocked = _mocked_modules(register_file_in_place=register_mock) + output = { + "images": [ + {"filename": "a.png", "subfolder": "", "type": "output"}, + {"filename": "b.png", "subfolder": "", "type": "output"}, + ] + } + + with patch.dict("sys.modules", mocked), \ + patch("os.path.isfile", return_value=True): + import importlib + import comfy_execution.asset_enrichment as mod + importlib.reload(mod) + mod.enrich_output_with_assets(output) + + self.assertEqual(register_mock.call_count, 2) + + def test_original_entry_not_mutated(self): + orig = {"filename": "a.png", "subfolder": "", "type": "output"} + output = {"images": [orig]} + _call(output) + self.assertNotIn("id", orig) + + def test_enrichment_error_does_not_block_sibling_entries(self): + call_count = [0] + good_reg = _make_register_result(ref_id="good-ref") + + def register_side_effect(abs_path, name, tags): + call_count[0] += 1 + if call_count[0] == 1: + raise RuntimeError("boom") + return good_reg + + mocked = _mocked_modules(register_file_in_place=register_side_effect) + + output = { + "images": [ + {"filename": "bad.png", "subfolder": "", "type": "output"}, + {"filename": "good.png", "subfolder": "", "type": "output"}, + ] + } + + with patch.dict("sys.modules", mocked), \ + patch("os.path.isfile", return_value=True): + import importlib + import comfy_execution.asset_enrichment as mod + importlib.reload(mod) + result = mod.enrich_output_with_assets(output) + + imgs = result["images"] + self.assertNotIn("id", imgs[0]) + self.assertEqual(imgs[1]["id"], "good-ref") + + def test_multiple_output_keys_all_enriched(self): + output = { + "images": [{"filename": "a.png", "subfolder": "", "type": "output"}], + "videos": [{"filename": "b.mp4", "subfolder": "", "type": "output"}], + } + result = _call(output) + self.assertIn("id", result["images"][0]) + self.assertIn("id", result["videos"][0]) + + def test_none_entry_in_list_unchanged(self): + output = {"images": [None, {"filename": "a.png", "subfolder": "", "type": "output"}]} + result = _call(output) + self.assertIsNone(result["images"][0]) + self.assertIn("id", result["images"][1]) + + def test_path_traversal_subfolder_skipped(self): + register_mock = MagicMock(return_value=_make_register_result()) + mocked = _mocked_modules(register_file_in_place=register_mock) + + output = {"images": [{"filename": "passwd", "subfolder": "../../etc", "type": "output"}]} + + # Do NOT patch os.path.abspath — real resolution is required for the containment check. + with patch.dict("sys.modules", mocked), \ + patch("os.path.isfile", return_value=True): + import importlib + import comfy_execution.asset_enrichment as mod + importlib.reload(mod) + result = mod.enrich_output_with_assets(output) + + self.assertNotIn("id", result["images"][0]) + register_mock.assert_not_called() + + def test_absolute_filename_skipped(self): + register_mock = MagicMock(return_value=_make_register_result()) + mocked = _mocked_modules(register_file_in_place=register_mock) + + # Absolute filename — os.path.join discards earlier components when a later one is absolute. + absolute_filename = os.path.abspath(os.sep + "etc" + os.sep + "passwd") + output = {"images": [{"filename": absolute_filename, "subfolder": "", "type": "output"}]} + + with patch.dict("sys.modules", mocked), \ + patch("os.path.isfile", return_value=True): + import importlib + import comfy_execution.asset_enrichment as mod + importlib.reload(mod) + result = mod.enrich_output_with_assets(output) + + self.assertNotIn("id", result["images"][0]) + register_mock.assert_not_called() + + +if __name__ == "__main__": + unittest.main() From 431a1888d31114ef4959c8a9fb286a5cac8688f0 Mon Sep 17 00:00:00 2001 From: Matt Miller Date: Wed, 10 Jun 2026 19:23:01 -0700 Subject: [PATCH 14/21] revert(assets): drop job_ids filter from GET /api/assets (#14408) The job_ids query filter added in #13998 has no live consumer: the frontend Generated tab kept sourcing from GET /jobs, and the cloud side removed its equivalent filter from the shared asset spec. Carrying it on the local server only re-introduces Core<->Cloud drift on the shared contract, so remove it to match. Removed: the job_ids field + validator on ListAssetsQuery, the IN(...) clauses in list_references_page, the service/route passthrough, and the filter-only tests. Kept: the canonical-UUID prompt_id enforcement at job creation (also landed in #13998). It stands on its own -- job ids are matched verbatim by history keys, websocket correlation, and /interrupt -- and cloud inherits it by running core for execution, so no divergence is created. --- app/assets/api/routes.py | 1 - app/assets/api/schemas_in.py | 36 ----------- .../database/queries/asset_reference.py | 6 -- app/assets/services/asset_management.py | 2 - comfy_execution/jobs.py | 7 +-- .../assets_test/queries/test_asset_info.py | 50 ---------------- .../queries/test_list_assets_query.py | 60 ------------------- .../assets_test/test_prompt_id_enforcement.py | 8 +-- tests/execution/test_jobs.py | 4 +- 9 files changed, 9 insertions(+), 165 deletions(-) delete mode 100644 tests-unit/assets_test/queries/test_list_assets_query.py diff --git a/app/assets/api/routes.py b/app/assets/api/routes.py index 6c9a3200d..7ef462f5c 100644 --- a/app/assets/api/routes.py +++ b/app/assets/api/routes.py @@ -219,7 +219,6 @@ async def list_assets_route(request: web.Request) -> web.Response: exclude_tags=q.exclude_tags, name_contains=q.name_contains, metadata_filter=q.metadata_filter, - job_ids=q.job_ids, limit=q.limit, offset=q.offset, sort=sort, diff --git a/app/assets/api/schemas_in.py b/app/assets/api/schemas_in.py index 4ae18c65a..af666746d 100644 --- a/app/assets/api/schemas_in.py +++ b/app/assets/api/schemas_in.py @@ -1,5 +1,4 @@ import json -import uuid from dataclasses import dataclass from typing import Any, Literal @@ -54,7 +53,6 @@ class ListAssetsQuery(BaseModel): include_tags: list[str] = Field(default_factory=list) exclude_tags: list[str] = Field(default_factory=list) name_contains: str | None = None - job_ids: list[str] = Field(default_factory=list, max_length=500) # Accept either a JSON string (query param) or a dict metadata_filter: dict[str, Any] | None = None @@ -88,40 +86,6 @@ class ListAssetsQuery(BaseModel): return out return v - @field_validator("job_ids", mode="before") - @classmethod - def _split_and_validate_job_ids(cls, v): - # Accept "uuid1,uuid2" or ["uuid1","uuid2"] or repeated query params. - # Each entry must parse as a UUID; canonicalized to lowercase hyphenated form. - if v is None: - return [] - if isinstance(v, str): - raw = [t.strip() for t in v.split(",") if t.strip()] - elif isinstance(v, list): - raw = [] - for item in v: - if not isinstance(item, str): - raise ValueError( - f"job_ids entries must be strings, got {type(item).__name__}" - ) - raw.extend([t.strip() for t in item.split(",") if t.strip()]) - else: - raise ValueError( - f"job_ids must be a string or list of strings, got {type(v).__name__}" - ) - - out: list[str] = [] - seen: set[str] = set() - for s in raw: - try: - canonical = str(uuid.UUID(s)) - except ValueError as e: - raise ValueError(f"job_ids must be UUIDs: {s!r}") from e - if canonical not in seen: - seen.add(canonical) - out.append(canonical) - return out - @field_validator("metadata_filter", mode="before") @classmethod def _parse_metadata_json(cls, v): diff --git a/app/assets/database/queries/asset_reference.py b/app/assets/database/queries/asset_reference.py index 33ded8a1c..792411800 100644 --- a/app/assets/database/queries/asset_reference.py +++ b/app/assets/database/queries/asset_reference.py @@ -264,7 +264,6 @@ def list_references_page( include_tags: Sequence[str] | None = None, exclude_tags: Sequence[str] | None = None, metadata_filter: dict | None = None, - job_ids: Sequence[str] | None = None, sort: str | None = None, order: str | None = None, after_cursor_value: object | None = None, @@ -294,9 +293,6 @@ def list_references_page( escaped, esc = escape_sql_like_string(name_contains) base = base.where(AssetReference.name.ilike(f"%{escaped}%", escape=esc)) - if job_ids: - base = base.where(AssetReference.job_id.in_(list(job_ids))) - base = apply_tag_filters(base, include_tags, exclude_tags) base = apply_metadata_filter(base, metadata_filter) @@ -349,8 +345,6 @@ def list_references_page( count_stmt = count_stmt.where( AssetReference.name.ilike(f"%{escaped}%", escape=esc) ) - if job_ids: - count_stmt = count_stmt.where(AssetReference.job_id.in_(list(job_ids))) count_stmt = apply_tag_filters(count_stmt, include_tags, exclude_tags) count_stmt = apply_metadata_filter(count_stmt, metadata_filter) diff --git a/app/assets/services/asset_management.py b/app/assets/services/asset_management.py index 53aec7a15..d4e4fc61c 100644 --- a/app/assets/services/asset_management.py +++ b/app/assets/services/asset_management.py @@ -274,7 +274,6 @@ def list_assets_page( exclude_tags: Sequence[str] | None = None, name_contains: str | None = None, metadata_filter: dict | None = None, - job_ids: Sequence[str] | None = None, limit: int = 20, offset: int = 0, sort: str = "created_at", @@ -320,7 +319,6 @@ def list_assets_page( exclude_tags=exclude_tags, name_contains=name_contains, metadata_filter=metadata_filter, - job_ids=job_ids, limit=fetch_limit, offset=offset, sort=sort, diff --git a/comfy_execution/jobs.py b/comfy_execution/jobs.py index 3fbcc3eb0..20ebae155 100644 --- a/comfy_execution/jobs.py +++ b/comfy_execution/jobs.py @@ -25,10 +25,9 @@ def validate_job_id(value) -> str: Job ids must be UUIDs in the canonical lowercase hyphenated form. The id is stored and compared verbatim everywhere downstream — history keys, - websocket events, /interrupt matching, and the assets ``job_ids`` filter - (a String(36) column matched exactly) — so accepting another spelling - would either rewrite the client's id behind its back or mint a job whose - outputs the filter can never find. Rejecting loudly beats both. + websocket events, and /interrupt matching — so accepting another spelling + would silently rewrite the client's id and then miss every exact-match + lookup. Rejecting loudly beats that. Returns the id unchanged. Raises ValueError when the value is not a string in canonical UUID form. diff --git a/tests-unit/assets_test/queries/test_asset_info.py b/tests-unit/assets_test/queries/test_asset_info.py index ba729a270..fe510e342 100644 --- a/tests-unit/assets_test/queries/test_asset_info.py +++ b/tests-unit/assets_test/queries/test_asset_info.py @@ -158,56 +158,6 @@ class TestListReferencesPage: refs, _, _ = list_references_page(session, sort="name", order="asc") assert refs[0].name == "large" - def test_job_ids_filter(self, session: Session): - asset = _make_asset(session, "hash1") - job_a = str(uuid.uuid4()) - job_b = str(uuid.uuid4()) - ref_a = _make_reference(session, asset, name="from_job_a") - ref_a.job_id = job_a - ref_b = _make_reference(session, asset, name="from_job_b") - ref_b.job_id = job_b - _make_reference(session, asset, name="no_job") - session.commit() - - # Single job filter - refs, _, total = list_references_page(session, job_ids=[job_a]) - assert total == 1 - assert refs[0].name == "from_job_a" - - # Multi-job filter (IN) - refs, _, total = list_references_page(session, job_ids=[job_a, job_b]) - names = sorted(r.name for r in refs) - assert total == 2 - assert names == ["from_job_a", "from_job_b"] - - # Unknown job id matches nothing - refs, _, total = list_references_page(session, job_ids=[str(uuid.uuid4())]) - assert total == 0 - assert refs == [] - - # Empty/None means no filter -> all three references - refs, _, total = list_references_page(session, job_ids=[]) - assert total == 3 - refs, _, total = list_references_page(session, job_ids=None) - assert total == 3 - - def test_job_ids_combined_with_other_filters(self, session: Session): - asset = _make_asset(session, "hash1") - job_a = str(uuid.uuid4()) - ref_match = _make_reference(session, asset, name="match.bin") - ref_match.job_id = job_a - ref_wrong_name = _make_reference(session, asset, name="other.bin") - ref_wrong_name.job_id = job_a - ref_wrong_job = _make_reference(session, asset, name="match.bin") - ref_wrong_job.job_id = str(uuid.uuid4()) - session.commit() - - refs, _, total = list_references_page( - session, job_ids=[job_a], name_contains="match" - ) - assert total == 1 - assert refs[0].id == ref_match.id - class TestFetchReferenceAssetAndTags: def test_returns_none_for_nonexistent(self, session: Session): diff --git a/tests-unit/assets_test/queries/test_list_assets_query.py b/tests-unit/assets_test/queries/test_list_assets_query.py deleted file mode 100644 index e8d3430e2..000000000 --- a/tests-unit/assets_test/queries/test_list_assets_query.py +++ /dev/null @@ -1,60 +0,0 @@ -"""Schema-level unit tests for ListAssetsQuery (no DB required).""" -import uuid - -import pytest -from pydantic import ValidationError - -from app.assets.api.schemas_in import ListAssetsQuery - - -class TestJobIdsValidator: - def test_csv_string_parses_and_canonicalizes(self): - a = "AAAAAAAA-BBBB-CCCC-DDDD-EEEEEEEEEEEE" - b = "11111111-2222-3333-4444-555555555555" - q = ListAssetsQuery.model_validate({"job_ids": f"{a},{b}"}) - # Canonicalized to lowercase - assert q.job_ids == [a.lower(), b] - - def test_repeated_query_params_as_list(self): - a = "11111111-1111-1111-1111-111111111111" - b = "22222222-2222-2222-2222-222222222222" - q = ListAssetsQuery.model_validate({"job_ids": [a, b]}) - assert q.job_ids == [a, b] - - def test_dedup_preserves_first_seen_order(self): - a = "11111111-1111-1111-1111-111111111111" - b = "22222222-2222-2222-2222-222222222222" - q = ListAssetsQuery.model_validate({"job_ids": [a, b, a]}) - assert q.job_ids == [a, b] - - def test_default_empty(self): - q = ListAssetsQuery.model_validate({}) - assert q.job_ids == [] - - def test_invalid_uuid_rejected(self): - with pytest.raises(ValidationError) as exc: - ListAssetsQuery.model_validate({"job_ids": "not-a-uuid"}) - assert "must be UUIDs" in str(exc.value) - - def test_non_string_list_item_rejected(self): - with pytest.raises(ValidationError) as exc: - ListAssetsQuery.model_validate( - {"job_ids": ["11111111-1111-1111-1111-111111111111", 42]} - ) - assert "must be strings" in str(exc.value) - - def test_non_string_non_list_value_rejected(self): - with pytest.raises(ValidationError) as exc: - ListAssetsQuery.model_validate({"job_ids": {"bad": "shape"}}) - assert "must be a string or list of strings" in str(exc.value) - - def test_max_length_enforced(self): - too_many = [str(uuid.uuid4()) for _ in range(501)] - with pytest.raises(ValidationError) as exc: - ListAssetsQuery.model_validate({"job_ids": too_many}) - assert exc.value.errors()[0]["type"] == "too_long" - - def test_max_length_boundary_accepted(self): - at_cap = [str(uuid.uuid4()) for _ in range(500)] - q = ListAssetsQuery.model_validate({"job_ids": at_cap}) - assert len(q.job_ids) == 500 diff --git a/tests-unit/assets_test/test_prompt_id_enforcement.py b/tests-unit/assets_test/test_prompt_id_enforcement.py index fb961beae..86a755c9f 100644 --- a/tests-unit/assets_test/test_prompt_id_enforcement.py +++ b/tests-unit/assets_test/test_prompt_id_enforcement.py @@ -1,9 +1,9 @@ """POST /prompt enforces canonical-UUID job ids at creation time. -Lives in assets_test because it uses this suite's booted-server fixture and -because the invariant exists for the assets pipeline: the GET /api/assets -``job_ids`` filter matches stored job ids exactly, so a job minted with a -non-canonical id would produce assets the filter can never find. +Lives in assets_test because it uses this suite's booted-server fixture. The +invariant itself is pipeline-wide: a job id is stored and compared verbatim +downstream — history keys, websocket correlation, and /interrupt matching — +so a job minted with a non-canonical id would miss every exact-match lookup. The prompt bodies here are intentionally invalid workflows — prompt_id validation happens before workflow validation, so a rejected id returns diff --git a/tests/execution/test_jobs.py b/tests/execution/test_jobs.py index 30e47071d..f7cb612e4 100644 --- a/tests/execution/test_jobs.py +++ b/tests/execution/test_jobs.py @@ -35,8 +35,8 @@ class TestValidateJobId: ) def test_non_canonical_spellings_rejected(self, variant): # uuid.UUID parses all of these, but accepting them would silently - # rewrite the client's id (history keys, websocket events, and the - # assets job_ids filter all match the stored form exactly). + # rewrite the client's id (history keys, websocket events, and + # /interrupt matching all match the stored form exactly). with pytest.raises(ValueError): validate_job_id(variant) From 74ee826790035be831c960e4c4bd60051273a99a Mon Sep 17 00:00:00 2001 From: Comfy Org PR Bot Date: Thu, 11 Jun 2026 12:15:53 +0900 Subject: [PATCH 15/21] chore(openapi): sync shared API contract from cloud@e3c52ad (#14406) --- openapi.yaml | 3 +++ 1 file changed, 3 insertions(+) diff --git a/openapi.yaml b/openapi.yaml index 58614103a..6e203b1cd 100644 --- a/openapi.yaml +++ b/openapi.yaml @@ -1067,6 +1067,9 @@ components: comfyui_version: description: ComfyUI version type: string + deploy_environment: + description: How this ComfyUI instance is deployed (e.g. cloud, local-git, local-portable, local-desktop) + type: string embedded_python: description: Whether using embedded Python type: boolean From 33e6ebd0d92b270e9bd79ea74e967f7e23e7d7e8 Mon Sep 17 00:00:00 2001 From: comfyanonymous <121283862+comfyanonymous@users.noreply.github.com> Date: Wed, 10 Jun 2026 20:27:08 -0700 Subject: [PATCH 16/21] I don't think this actually works anymore. (#14403) --- README.md | 10 ---------- 1 file changed, 10 deletions(-) diff --git a/README.md b/README.md index dc2389266..ee1024de5 100644 --- a/README.md +++ b/README.md @@ -462,16 +462,6 @@ To use the most up-to-date frontend version: This approach allows you to easily switch between the stable fortnightly release and the cutting-edge daily updates, or even specific versions for testing purposes. -### Accessing the Legacy Frontend - -If you need to use the legacy frontend for any reason, you can access it using the following command line argument: - -``` ---front-end-version Comfy-Org/ComfyUI_legacy_frontend@latest -``` - -This will use a snapshot of the legacy frontend preserved in the [ComfyUI Legacy Frontend repository](https://github.com/Comfy-Org/ComfyUI_legacy_frontend). - # QA ### Which GPU should I buy for this? From bda19b26048843f32ef41e94f83b4ef49d16c254 Mon Sep 17 00:00:00 2001 From: rattus <46076784+rattus128@users.noreply.github.com> Date: Thu, 11 Jun 2026 13:32:57 +1000 Subject: [PATCH 17/21] ops: tolerate already force casted dynamic weight (#14410) Some custom nodes .to weights completely out of load context which can wreak havoc if its for a model that is not active. Detect this condition and just let it fall-through to the non-dynamic loader straight up. --- comfy/ops.py | 16 ++++++++-------- 1 file changed, 8 insertions(+), 8 deletions(-) diff --git a/comfy/ops.py b/comfy/ops.py index 119177c37..3c9912aae 100644 --- a/comfy/ops.py +++ b/comfy/ops.py @@ -299,21 +299,21 @@ def cast_bias_weight(s, input=None, dtype=None, device=None, bias_dtype=None, of non_blocking = comfy.model_management.device_supports_non_blocking(device) - if hasattr(s, "_v"): + if hasattr(s, "_v") and comfy.model_management.is_device_cpu(device): #vbar doesn't support CPU weights, but some custom nodes have weird paths #that might switch the layer to the CPU and expect it to work. We have to take #a clone conservatively as we are mmapped and some SFT files are packed misaligned #If you are a custom node author reading this, please move your layer to the GPU #or declare your ModelPatcher as CPU in the first place. - if comfy.model_management.is_device_cpu(device): - materialize_meta_param(s, ["weight", "bias"]) - weight = s.weight.to(dtype=dtype, copy=True) - if isinstance(weight, QuantizedTensor): - weight = weight.dequantize() - bias = s.bias.to(dtype=bias_dtype, copy=True) if s.bias is not None else None - return format_return((weight, bias, (None, None, None)), offloadable) + materialize_meta_param(s, ["weight", "bias"]) + weight = s.weight.to(dtype=dtype, copy=True) + if isinstance(weight, QuantizedTensor): + weight = weight.dequantize() + bias = s.bias.to(dtype=bias_dtype, copy=True) if s.bias is not None else None + return format_return((weight, bias, (None, None, None)), offloadable) + elif hasattr(s, "_v") and s.weight.device != device: prefetched = hasattr(s, "_prefetch") offload_stream = None offload_device = None From 91187c58d946f237a051a98eb3ef3ccd28cf926f Mon Sep 17 00:00:00 2001 From: Barish Ozbay <17261091+drozbay@users.noreply.github.com> Date: Wed, 10 Jun 2026 23:37:43 -0600 Subject: [PATCH 18/21] Improve context window resizing for SCAIL2 (CORE-286) (#14394) --- comfy/model_base.py | 19 ++++++++++++++++++- 1 file changed, 18 insertions(+), 1 deletion(-) diff --git a/comfy/model_base.py b/comfy/model_base.py index 2289e0812..ab4a11022 100644 --- a/comfy/model_base.py +++ b/comfy/model_base.py @@ -1816,7 +1816,24 @@ class WAN21_SCAIL2(WAN21_SCAIL): def resize_cond_for_context_window(self, cond_key, cond_value, window, x_in, device, retain_index_list=[]): if cond_key in ("sam_latents", "pose_latents"): - return comfy.context_windows.slice_cond(cond_value, window, x_in, device, temporal_dim=2, temporal_offset=1) + # Return sliced view omitting retain_index_list + return comfy.context_windows.slice_cond(cond_value, window, x_in, device, temporal_dim=2, temporal_offset=0) + if cond_key == "ref_mask_latents" and hasattr(cond_value, "cond") and isinstance(cond_value.cond, torch.Tensor): + # The ref mask is just a single frame padded with frames of zeros, so just grab the first frames for all windows + full_ref_mask = cond_value.cond + video_frame_count = x_in.shape[2] + if full_ref_mask.shape[2] != video_frame_count + 1: + return None + window_length = len(window.index_list) + + # Account for the causal anchor frame if it exists + anchor_index = getattr(window, "causal_anchor_index", None) + if anchor_index is not None and anchor_index >= 0: + window_length += 1 + + window_ref_mask = full_ref_mask[:, :, :window_length + 1].to(device) + return cond_value._copy_with(window_ref_mask) + return super().resize_cond_for_context_window(cond_key, cond_value, window, x_in, device, retain_index_list=retain_index_list) def concat_cond(self, **kwargs): From b97e60fc6beb795adaae404f5d43101dacf12f6f Mon Sep 17 00:00:00 2001 From: =?UTF-8?q?Jukka=20Sepp=C3=A4nen?= <40791699+kijai@users.noreply.github.com> Date: Thu, 11 Jun 2026 11:17:04 +0300 Subject: [PATCH 19/21] Fix SCAIL-2 reference mask background convention (#14415) --- comfy_extras/nodes_scail.py | 10 +++++++--- 1 file changed, 7 insertions(+), 3 deletions(-) diff --git a/comfy_extras/nodes_scail.py b/comfy_extras/nodes_scail.py index a740442de..bba0942d7 100644 --- a/comfy_extras/nodes_scail.py +++ b/comfy_extras/nodes_scail.py @@ -267,7 +267,8 @@ class SCAIL2ColoredMask(io.ComfyNode): io.Combo.Input("sort_by", options=["none", "left_to_right", "area"], default="left_to_right", tooltip="Order in which palette colors are assigned to the tracked objects (applied to both reference and pose video so each identity keeps the same color). left_to_right = leftmost object (by first-frame centroid) gets the first color; area = biggest object (by first-frame mask area) gets the first color; none = keep SAM3's order."), io.Boolean.Input("replacement_mode", default=False, - tooltip="False = mask_video has black bg (Animation Mode). True = white bg (Replacement Mode). Set the matching replacement_mode on WanSCAILToVideo. reference_image_mask is always black-bg regardless."), + tooltip="False = Animation Mode (pose_video_mask has black background, reference_image_mask has white background). " + "True = Replacement Mode (pose_video_mask has white background, reference_image_mask has black background)."), ], outputs=[ io.Image.Output("pose_video_mask"), @@ -296,14 +297,17 @@ class SCAIL2ColoredMask(io.ComfyNode): return td drv = _prep(driving_track_data) + # Animation: driving=black, ref=white. Replacement: driving=white, ref=black. mask_video = _render_colored_masks(drv, "white" if replacement_mode else "black") + ref_bg = "black" if replacement_mode else "white" if ref_track_data is not None: ref = _prep(ref_track_data) - reference_image_mask = _render_colored_masks(ref, "black") + reference_image_mask = _render_colored_masks(ref, ref_bg) else: H, W = drv["orig_size"] - reference_image_mask = torch.zeros(1, H, W, 3, device=comfy.model_management.intermediate_device(), dtype=comfy.model_management.intermediate_dtype()) + fill_value = 1.0 if ref_bg == "white" else 0.0 + reference_image_mask = torch.full((1, H, W, 3), fill_value, device=comfy.model_management.intermediate_device(), dtype=comfy.model_management.intermediate_dtype()) return io.NodeOutput(mask_video, reference_image_mask) From ef470b61e4eab7de3319a83e689a9f236138102f Mon Sep 17 00:00:00 2001 From: Alexander Piskun <13381981+bigcat88@users.noreply.github.com> Date: Thu, 11 Jun 2026 11:28:17 +0300 Subject: [PATCH 20/21] [Partner Nodes] fix(GPT Image): handle mismatched image sizes returned when size="auto" (#14414) Signed-off-by: bigcat88 --- comfy_api_nodes/nodes_openai.py | 12 +++++++++++- 1 file changed, 11 insertions(+), 1 deletion(-) diff --git a/comfy_api_nodes/nodes_openai.py b/comfy_api_nodes/nodes_openai.py index 0fe5fb9d0..ad62f2164 100644 --- a/comfy_api_nodes/nodes_openai.py +++ b/comfy_api_nodes/nodes_openai.py @@ -9,6 +9,7 @@ from PIL import Image from typing_extensions import override import folder_paths +from comfy.utils import common_upscale from comfy_api.latest import IO, ComfyExtension, Input from comfy_api_nodes.apis.openai import ( InputFileContent, @@ -62,7 +63,8 @@ async def validate_and_cast_response(response, timeout: int = None) -> torch.Ten timeout: Request timeout in seconds. Defaults to None (no timeout). Returns: - A torch.Tensor representing the image (1, H, W, C). + A torch.Tensor of shape (N, H, W, C) with all returned images; images whose + dimensions differ from the first image's are resized to match it. Raises: ValueError: If the response is not valid. @@ -89,6 +91,14 @@ async def validate_and_cast_response(response, timeout: int = None) -> torch.Ten arr = np.asarray(pil_img).astype(np.float32) / 255.0 image_tensors.append(torch.from_numpy(arr)) + # With size="auto" the API can return images whose dimensions differ by a few pixels within a single response + # resize them to the first image's dimensions so they can be stacked into one batch. + ref_h, ref_w = image_tensors[0].shape[:2] + for i, t in enumerate(image_tensors): + if t.shape[:2] != (ref_h, ref_w): + samples = t.unsqueeze(0).movedim(-1, 1) + samples = common_upscale(samples, ref_w, ref_h, "bilinear", "center") + image_tensors[i] = samples.movedim(1, -1).squeeze(0) return torch.stack(image_tensors, dim=0) From fb991e2c1e7476809d566a4620c2132e05a466dd Mon Sep 17 00:00:00 2001 From: Alexander Piskun <13381981+bigcat88@users.noreply.github.com> Date: Thu, 11 Jun 2026 17:43:35 +0300 Subject: [PATCH 21/21] [Partner Nodes] fix(KlingTextToVideoNode): validation error for "kling-v2-master" model (#14418) Signed-off-by: bigcat88 --- comfy_api_nodes/apis/__init__.py | 9 +-------- comfy_api_nodes/nodes_kling.py | 2 +- 2 files changed, 2 insertions(+), 9 deletions(-) diff --git a/comfy_api_nodes/apis/__init__.py b/comfy_api_nodes/apis/__init__.py index 9c4cfb9b6..9a7049ea2 100644 --- a/comfy_api_nodes/apis/__init__.py +++ b/comfy_api_nodes/apis/__init__.py @@ -1310,13 +1310,6 @@ class KlingTaskStatus(str, Enum): failed = 'failed' -class KlingTextToVideoModelName(str, Enum): - kling_v1 = 'kling-v1' - kling_v1_6 = 'kling-v1-6' - kling_v2_1_master = 'kling-v2-1-master' - kling_v2_5_turbo = 'kling-v2-5-turbo' - - class KlingVideoGenAspectRatio(str, Enum): field_16_9 = '16:9' field_9_16 = '9:16' @@ -5179,7 +5172,7 @@ class KlingText2VideoRequest(BaseModel): duration: Optional[KlingVideoGenDuration] = '5' external_task_id: Optional[str] = Field(None, description='Customized Task ID') mode: Optional[KlingVideoGenMode] = 'std' - model_name: Optional[KlingTextToVideoModelName] = 'kling-v1' + model_name: Optional[str] = 'kling-v1' negative_prompt: Optional[str] = Field( None, description='Negative text prompt', max_length=2500 ) diff --git a/comfy_api_nodes/nodes_kling.py b/comfy_api_nodes/nodes_kling.py index d11e42540..c81d3503d 100644 --- a/comfy_api_nodes/nodes_kling.py +++ b/comfy_api_nodes/nodes_kling.py @@ -436,7 +436,7 @@ async def execute_text2video( negative_prompt=negative_prompt if negative_prompt else None, duration=KlingVideoGenDuration(duration), mode=KlingVideoGenMode(model_mode), - model_name=KlingVideoGenModelName(model_name), + model_name=model_name, cfg_scale=cfg_scale, aspect_ratio=KlingVideoGenAspectRatio(aspect_ratio), camera_control=camera_control,